Professional Documents
Culture Documents
Livestock Ration Formulation For Dairy Cattle and Buffalo CRC Press
Livestock Ration Formulation For Dairy Cattle and Buffalo CRC Press
me/livestockscience
Livestock Ration
Formulation for Dairy
Cattle and Buffalo
e
nc
ie
sc
k
oc
st
ve
li
e/
t.m
s://
tp
ht
https://t.me/livestockscience
e
nc
ie
sc
ck
to
es
liv
e/
.m
//t
s:
tp
ht
https://t.me/livestockscience
Livestock Ration
Formulation for Dairy
Cattle and Buffalo
Radha Gupta
Department of Mathematics, Dayananda Sagar College of Engineering,
Bengaluru, Karnataka, India
D. Rajendran
ICAR-National Institute of Animal Nutrition and Physiology, Bengaluru,
Karnataka, India
Vishal Patil
Department of Mathematics, Faculty of Engineering and Technology, Jain
(Deemed-to-be) University, Bengaluru, Karnataka, India
https://t.me/livestockscience
MATLAB® is a trademark of The MathWorks, Inc. and is used with permission. The
MathWorks does not warrant the accuracy of the text or exercises in this book. This
book’s use or discussion of MATLAB® software or related products does not constitute
endorsement or sponsorship by The MathWorks of a particular pedagogical approach or
particular use of the MATLAB® software.
First edition published 2022
by CRC Press
6000 Broken Sound Parkway NW, Suite 300, Boca Raton, FL 33487–2742
and by CRC Press
2 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN
© 2022 Ravinder Singh Kuntal, Radha Gupta, Rajendran D and Vishal Patil
CRC Press is an imprint of Taylor & Francis Group, LLC
Reasonable efforts have been made to publish reliable data and information, but the
author and publisher cannot assume responsibility for the validity of all materials or the
consequences of their use. The authors and publishers have attempted to trace the copyright
holders of all material reproduced in this publication and apologize to copyright holders if
permission to publish in this form has not been obtained. If any copyright material has not
been acknowledged please write and let us know so we may rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted,
reproduced, transmitted, or utilized in any form by any electronic, mechanical, or other means,
now known or hereafter invented, including photocopying, microflming, and recording, or in
any information storage or retrieval system, without written permission from the publishers.
For permission to photocopy or use material electronically from this work, access www.
copyright.com or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood
Drive, Danvers, MA 01923, 978–750–8400. For works that are not available on CCC please
contact mpkbookspermissions@tandf.co.uk
Trademark notice: Product or corporate names may be trademarks or registered trademarks
and are used only for identifcation and explanation without intent to infringe.
Library of Congress Cataloging‑in‑Publication Data
Names: Singh Kuntal, Ravinder, author. | Gupta, Radha, 1970– author. |
Rajendran, D., author. | Patil, Vishal, author.
Title: Livestock ration formulation for dairy cattle and buffalo /
Ravinder Singh Kuntal, Radha Gupta, Rajendran D, Vishal Patil.
Description: First edition | Boca Raton, FL : CRC Press, 2022. |
Includes bibliographical references and index.
Identifers: LCCN 2021048758 | ISBN 9781032137476 (hardback) |
ISBN 9781032139678 (paperback) | ISBN 9781003231714 (ebook)
Subjects: LCSH: Dairy cattle—Feeds and feeding. | Buffaloes—Feeds and
feeding. | Dairy farming. | Animal nutrition. | Handbooks and manuals.
Classifcation: LCC SF95 .S564 2022 | DDC 636.08/52—dc23/eng/20211122
LC record available at https://lccn.loc.gov/2021048758
ISBN: 978-1-032-13747-6 (hbk)
ISBN: 978-1-032-13967-8 (pbk)
ISBN: 978-1-003-23171-4 (ebk)
DOI: 10.1201/9781003231714
Typeset in Times
by Apex CoVantage, LLC
https://t.me/livestockscience
This book is dedicated to the memory of my beloved father
&
To my mother
This book was written when you were separated from me,
https://t.me/livestockscience
e
nc
ie
sc
ck
to
es
liv
e/
.m
//t
s:
tp
ht
https://t.me/livestockscience
Contents
List of Tables........................................................................................................... xi
List of Figures.......................................................................................................xiii
Preface ................................................................................................................... xv
Acknowledgments................................................................................................xvii
Authors..................................................................................................................xix
List of Abbreviations ..........................................................................................xxiii
vii
https://t.me/livestockscience
viii Contents
https://t.me/livestockscience
Contents ix
7. Conclusion..................................................................................................... 87
References............................................................................................................. 95
Index .................................................................................................................... 103
https://t.me/livestockscience
e
nc
ie
sc
ck
to
es
liv
e/
.m
//t
s:
tp
ht
https://t.me/livestockscience
Tables
Table 1.1 Nutrient Content of Common Feeds and Fodder for Dairy
Cattle and Buffalo on DM Basis ........................................................ 6
Table 4.1 Composition of Concentrate Mixtures in Respect of DCP
and TDN ........................................................................................... 28
Table 4.2 Solution for NLP Model by GA for Original Bounds
(Test Problem-1) ............................................................................... 30
Table 4.3 Solution for NLP Model by GA for Reduced Bounds
(Test Problem-2) ............................................................................... 31
Table 4.4 Comparison of Solution Obtained by GA and RST for
NLP Model ....................................................................................... 31
Table 4.5 Nonlinear GP Model with Priority Ranked Goals for Animal
at Level-1 to 3 with Objective Function (z) to Different Goals
Which Need to Be Minimized ......................................................... 35
Table 4.6 Solution Set of LPP for Level-1, 2 and 3 by GA .............................. 36
Table 4.7 Solution Set of the Nonlinear Programming Problem (NLPP)
for Level-1, 2 and 3 by GA................................................................ 37
Table 4.8 Comparison of Decision Variables by RST and GA........................ 38
Table 5.1 Nutrient Requirements of Cattle-1, 2 and 3 ..................................... 47
Table 5.2 Selected Feedstuffs........................................................................... 47
Table 5.3 Nutrient Content of Feedstuffs on DM Basis................................... 48
Table 5.4 Least Cost of Ration by Various Techniques for Cattle-1
on DM Basis ..................................................................................... 54
Table 5.5 Least Cost of Ration by Various Techniques for Cattle-1
on As Fresh Basis (LP Model) ......................................................... 54
Table 5.6 Least Cost of Ration by Various Techniques for Cattle-2
on DM Basis (LP Model) ................................................................. 55
Table 5.7 Least Cost of Ration by Various Techniques for Cattle-2
on As Fresh Basis (LP Model) ......................................................... 56
Table 5.8 Least Cost of Ration by Various Techniques for Cattle-3
on DM Basis (LP Model) ................................................................. 56
xi
https://t.me/livestockscience
xii Tables
https://t.me/livestockscience
Figures
xiii
https://t.me/livestockscience
e
nc
ie
sc
ck
to
es
liv
e/
.m
//t
s:
tp
ht
https://t.me/livestockscience
Preface
xv
https://t.me/livestockscience
https://t.me/livestockscience
Acknowledgments
“I bow to Goddess Saraswati, who fulfls the wishes of the devotees. I pray to her
to enlighten me with knowledge. Saraswati is the provider of boons and the one
who grants all our desires. She is capable of removing ignorance and bestowing
intelligence. I pray to the Goddess to help me in making this work fruitful and
make me successful in all my efforts.” I am grateful to Lord Ganesha for giving me
patience and strength to overcome diffculties, which helped me crossing my way
in the accomplishment of this endeavour.
It is hard to say whether it has been grappling with the topic itself which has been
the real learning experience or grappling with how to write papers, give talks, work
in a group, stay up until the birds start singing and stay focused. In any case, I am
indebted to many people for making the time working on my book an unforgettable
experience. It would not have been possible to write this book without the help and
support of the kind people around me. I acknowledge many known and unknown
hands that have helped me in this process. No words could adequately express my
gratitude to them, and I shall ever remain indebted to them all. I express my heart
with a profound sense of gratitude to my supervisor Dr. Radha Gupta, Professor
and Head, Department of Mathematics, Dayananda Sagar College of Engineering,
Bangalore. She has oriented and supported me with promptness and care and has
always been patient and encouraging in times of new ideas and diffculties. Her
excellent quality in teaching the subject will remain in my mind forever. Her hard
work, constant patience and determined mind set an example. This book would
not have been possible without her help, constant support and patience. Her invalu-
able help of constructive comments and suggestions throughout my research work
have contributed to the progress of this research. I am very much thankful to
Dr. D. Rajendran, Principal Scientist, Animal Nutrition Division, National Institute
of Animal Nutrition and Physiology (NIANP), Adugodi, Bangalore, for provid-
ing primary data and continuous guidance in this research journey. His enormous
research experience as well as subject knowledge is source of inspiration for me
which helped me to carry this work to this level. He also helped me in improv-
ing my writing skills, fnding appropriate journals and in making an Excel-based
nutrient bound calculator for calculating nutrient requirements of dairy cattle and
buffaloes and result validation. Above all, he made me feel as a friend, which I
appreciate from my heart. I am deeply grateful to Dr. Sandeep Shastri, Professor
and the Vice Chancellor of Vice Chancellor, Jagran Lake City University, Bhopal,
who has made me understand the word research. His innovative and systematic
way of teaching research methodology has helped me to carry out the work in an
effective way. He has provided the right direction to the entire path of research and
xvii
https://t.me/livestockscience
xviii Acknowledgments
enlightened and entertained us with his knowledge, experience and skill in this
feld. I can’t imagine my current position without the support from my mother and
sister. A formal statement of acknowledgment will hardly meet the end of justice
in the matter of the expression of a deep sense of gratitude to my mother, Smt.
Premvati Devi. I express my deep sense of love and gratitude for her great patience
at all times, constant encouragement and inspiration. I also express my deep sense
of gratitude to Dr. Anita Chaturvedi, Professor, Department of Mathematics, FET
Jain (Deemed-to-be) University for her kind support and encouragement at all times.
I am also very much thankful to Dr. Kokila Ramesh, Department of Mathematics,
FET Jain (Deemed-to-be) University for extending her help in times of need. I
also express my deep sense of gratitude to Ms. Priya Chakraborty, Department of
English, FET Jain (Deemed-to-be) University for extending her help in correcting
grammatical errors in this text book.
https://t.me/livestockscience
Authors
xix
https://t.me/livestockscience
xx Authors
https://t.me/livestockscience
Authors xxi
was recognized and received best thesis award by the Animal Nutrition Journal by
Elsevier Pvt. Ltd.
He has 185 publications, including 51 research articles, 12 book chapters, 37
technical articles and 20 teaching manuals. He received seven awards in con-
ferences and seminars of national and international repute. He was invited as
guest lecturer by CG Group at Boun Me Thout, DakLak, Vietnam. He success-
fully guided 12 master degree and fve doctoral degree students in the capacity
of chairman/member of the advisory committee. He developed four software and
two Android apps and published them for public use. He has published fve ready
reckoners and one diagnostic kit. More than 25000 copies of the ready reckoner
were sold and created a revolution in the feld of animal nutrition. He has success-
fully completed 11 externally/institute funded research projects, and currently has
fve projects in progress. The International Society of Biotechnology conferred the
Fellow of International Society of Biotechnology (FISBT) on him in recognition of
his contribution and achievements.
https://t.me/livestockscience
https://t.me/livestockscience
Abbreviations
b Bran Max
BGA Binary Coded Genetic Algorithm
Ca Calcium
c/h Legume/Non-Legume Green
Ck Cake Max
CP Crude Protein
d/g Dry/Green Roughage
DM Dry Matter
DMI Dry Matter Intake
EA Evolutionary Algorithm
GA Genetic Algorithm
gm/kg Gram per Kilogram
GP/GPP Goal Programming/Goal Programming Problem
GRG Generalized Reduced Gradient
ICAR Indian Council of Agricultural Research
INR Indian Rupees
KT Kuhn-Tucker
lc Least Cost
LP/LPP Linear Programming/Linear Programming Problem
NIANP National Institute of Animal Nutrition and Physiology
NLP/NLPP Nonlinear Programming/ Nonlinear Programming Problem
NRC National Research Council
P Phosphorus
r/c Roughage/Concentrate
RGA Real Coded Genetic Algorithm
RHGA Real Coded Hybrid Genetic Algorithm
RNG Random Number Generation
RST Controlled Random Search Technique
SQP Sequential Quadratic Programming
t Total (in Kg)
TDN Total Digestible Nutrients
TMR Total Mixed Ration
xxiii
https://t.me/livestockscience
e
nc
ie
sc
ck
to
es
liv
e/
.m
//t
s:
tp
ht
https://t.me/livestockscience
1
Introduction to Animal Nutrition
CONTENTS
1.1 Operational Research ..................................................................................... 1
1.2 Animal Nutrition ............................................................................................ 2
1.3 Essential Nutrients ......................................................................................... 3
1.3.1 Energy................................................................................................ 4
1.3.2 Protein ............................................................................................... 4
1.3.3 Minerals............................................................................................. 4
1.4 Classifcation of Feedstuffs ............................................................................ 5
1.5 Application Area ............................................................................................ 6
Yet this explicit defnition is not suffcient, as it is different from a related feld.
However, operations research is a subfeld of applied mathematics. We are using
techniques from mathematics such as mathematical modeling, statistical analy-
sis and mathematical optimization to arrive at optimal or near-optimal solutions
to complex problems. Therefore, instead of following the INFORMS defnition,
we use operations research with the help of mathematical optimization to specify
DOI: 10.1201/9781003231714-1 1
https://t.me/livestockscience
2 Livestock Ration Formulation
techniques and solutions for specifc real-time problems. This method includes lin-
ear programming, nonlinear programming, integer programming, optimization,
the Markov process, queuing theory, goal programming and heuristic algorithms.
These techniques would be accepted because they give optimal solutions for spe-
cifc problems except for the fact that in linear programming, if the problem is not
defned in linear equations (equalities or inequalities) or continuous domain, then
we cannot apply linear programming algorithms. There are many techniques in
operations research which have proven to be very useful in practice. For many real-
world optimization problems, operations research provides the method of choice,
in particular for those problems that are complex in nature.
https://t.me/livestockscience
Introduction to Animal Nutrition 3
addition, during 2016–17, the share of milk production from cross-breed cows,
indigenous cows, nondescript cows and buffaloes was 25.4%, 11.3%, 9.5% and
49.2%, respectively, of total milk production. The increase in population of exotic/
cross-breed cattle and indigenous cattle was 34.8 % and 0.17%, respectively,
whereas the population of milk animals (cows and buffaloes) has increased from
77.04 million to 80.52 million. Increasing the productivity of animals is one of the
major issues in our country, which can be overcome by developing proper feeding
systems to provide balanced nutrients to fulfll their requirements. By looking at
increasing demands of animal products such as milk and dairy products, we can
see a growth in a new entrepreneurship area with a commercial outlook for which
scientifc management and sustainability are important components. Nutrients
are very important to dairy animals to maintain their body conditions, for milk
production with an adequate amount of fat percentage and for maintaining preg-
nancy at third trimester. These nutrients are provided through feed, and the nutrient
requirements are measured in terms of energy, protein, minerals and vitamins. In
many developed countries, the standard feeding practice has come through various
continuous experiments, evaluation and refnement for decades. But in India, the
frst published document on scientifc feeding for Indian cattle appeared in ICAR
bulletin No 25, titled “Nutritive Values of Indian Cattle Feeds and the Feeding of
Animals” by Sen (1957). This was later modifed by Sen and Ray (1964), and these
feeding practices was revised by Sen, Ray and Ranjhan (1978) with some research
on Indian breeds. Kearl in 1982 compiled data on nutritional requirements of dif-
ferent livestock species and values of different feedstuffs from various developing
countries in a systematic manner. After reviewing these facts, India compiled these
standards, and as a result, nutritional requirements of cattle and buffalo were pub-
lished by ICAR in 1985 and 1998. After this publication, a change in productivity
of Indian livestock and nutritive values were experienced in new verity of crop, and
modern agriculture practices were introduced, which allows new methodologies for
animal nutrition. In 2013, ICAR came out with nutrient requirement of cattle and
buffaloes and other species to enhance their productivity. Unfortunately, feeding
standards have not been popularized to all dairy farmers due to lack of aware-
ness among farmers as well as lack of resources. In fact, there is a need to provide
knowledge of feeding practice to farmers; though there are many government bod-
ies working on these issues in rural areas, there is still need to enhance the overall
productivity. Therefore, in present research, an effort has been made by to provide
a least cost balanced ration to livestock at different conditions.
https://t.me/livestockscience
4 Livestock Ration Formulation
1.3.1 Energy
Per se, energy is not a nutrient, but it is an important component for all types of
body functions which needs to be supplied through diets because dairy cattle and
buffalos need to maintain their energy level for body maintenance, reproduction,
lactation and growth and during pregnancy. Energy requirement for dairy animal
are expressed as total digestible nutrients (TDN), where the requirement of TDN
will differ according to animal physiological condition.
1.3.2 Protein
Protein is basically made up of amino acids by peptide bonds, which are used
to make tissues. Protein is important for growth and milk production and to
maintain body conditions. Livestock use amino acids from digested protein to
repair and to build tissues and also utilize microbial protein produced in rumen
during the protein degradation process. Because of rumen microbes in cattle,
rumen can break down about 80% of protein in the feed and form microbial
protein. Protein requirements for dairy animals usually are expressed as crude
protein (CP).
1.3.3 Minerals
Minerals are important for various body functions of dairy animals. Minerals,
along with proteins, form bone and teeth, and some minerals help in nerve
impulse transmission and to carry oxygen. These are basically two types – macro
and micro. For dairy animals, the major minerals are macro minerals such as
calcium, phosphorus, magnesium, sodium, potassium, chloride and sulphur
which is required in diet at a higher level, which is measured as a percentage of
dry matter. Based on different physiological conditions of livestock, minerals
must be included for optimum performance and health. Therefore, by consider-
ing these facts, the required value of DMI, TDN and CP for dairy animals has
been calculated as per the ICAR 2013 standards, while values of calcium and
https://t.me/livestockscience
Introduction to Animal Nutrition 5
https://t.me/livestockscience
6 Livestock Ration Formulation
TABLE 1.1
Nutrient Content of Common Feeds and Fodder for Dairy Cattle and Buffalo on DM
Basis
Name of the ingredients DM % CP % TDN % Ca % P%
Green roughage
Bajra Napier CO-4 grass 25 8 52 0.144 0.09
Maize fodder 25 8 60 0.53 0.14
Sorghum Co-FS 29 fodder 90 7 50 0.12 0.09
Berseem 90 5 45 0.65 0.05
Dry roughage
Paddy straw 90 5.13 40 0.18 0.08
Maize stover 90 3 42 0.145 0.027
Ragi straw 90 6 42 0.15 0.09
Concentrate
Maize 90 8.1 79.2 0.018 0.27
Soya DOC 90 42 70 0.018 0.225
Copra DOC 90 22 70 0.036 0.9
Cotton DOC 90 32 70 0.036 0.9
Wheat bran 75 12 70 1.067 0.093
Gram chunnies 90 17 60 0.108 0.234
Cotton seed 90 16 110 0.3 0.62
Chickpea husk 90 18 45 0.3 0.62
Concentrate mix Type I 90 22 70 0.5 0.45
Concentrate mix Type II 90 20 65 0.5 0.4
Minerals
Calcite 97 0 0 36 0
Grit 96 0 0 36 0
MM 90 0 0 32 6
DCP 90 0 0 24 16
Soda bicarbonate 90 0 0 0 0
Salt 90 0 0 0 0
TM mix 98 0 0 0 0
Urea 95 287.5 0 0 0
https://t.me/livestockscience
Introduction to Animal Nutrition 7
https://t.me/livestockscience
https://t.me/livestockscience
2
Review of Literature
CONTENTS
2.1 Introduction .................................................................................................... 9
2.2 Linear Programming in Ration Formulation ............................................... 10
2.2.1 Linear Programming Based Software............................................. 13
2.2.2 Limitations of Linear Programming in Ration Formulation .......... 13
2.2.3 Overcoming the Limitations of Linear Programming .................... 14
2.3 Nonlinear Programming .............................................................................. 14
2.4 Goal Programming ...................................................................................... 18
2.5 Genetic Algorithm ....................................................................................... 19
2.1 Introduction
Feed formulation is the process of quantifying the amounts of different feed
ingredients that need to be put together to form a single uniform mixture (diet)
for animals that supplies all their nutrient requirements. It is concerned with the
allocation of nutrient ingredients to the animal in terms of yield and weight gain at
the least possible cost. It is one of the central operations of the animal industry, as
the performance and the optimal growth of animals is entirely dependent on their
feed intake. To meet the animals’ requirement at a particular stage of production,
it is very important to formulate the diet effciently with low cost, as feed costs
account for 40–80% of the total costs in animal production (Pond et al., 1995),
75–85% in pig production (Gallenti, 1997: 15–19) and over 60% in poultry pro-
duction (Rose, 1997), and in milk production feed costs are the largest expense
(Bath, 1985: 1579–1584). Therefore, procedures that reduce feed costs are likely
to increase net incomes in agriculture. Feed formulation is one of the areas that
one can use to reduce the cost of feed (Nabasirye et al., 2011: 221–226). Rations
for livestock should be formulated in such a way that they should supply all essen-
tial nutrients and energy to maintain the vital physiological functions of growth,
reproduction and health of animals. To formulate the low-cost diet, planning the
linear programming-based diet formulation has come into existence with the help
of mathematicians. This method is widely used in commercial dairy farms and
commercially available software. The least cost formulation of rations based on
linear programming optimizes the combination of feed ingredients which supplies
the required amount of nutrients at least cost (Rossi, 2004). The major nutrients to
DOI: 10.1201/9781003231714-2 9
https://t.me/livestockscience
10 Livestock Ration Formulation
be supplied in a diet are energy, protein and minerals such as calcium and phos-
phorus for maintenance, growth, pregnancy and milk production of dairy animals.
These nutrients should be supplied in required quantity to meet the specifc per-
formance targets. These entire concepts need a professional knowledge of animal
nutrition to consider the requirements of the nutrients and animals’ capability to
digest and then assimilate the nutrients from various available ingredients.
Livestock ration formulation models are developed for commercial purposes as
well as for development, using various mathematical techniques for several decades.
Some of them use the Pearson’s square method: the Pearson’s square ration formu-
lation procedure is designed for simple rations and has been used for many years.
It is direct and easy to follow and is useful in balancing protein requirements. The
limitation of this technique is that it satisfes only one nutrient requirement and uses
only two feed ingredients. It therefore has limited application where farmers must
formulate diets balanced for protein, energy, minerals and least cost (Chakeredza
et al., 2008: 2925–2933). The simultaneous equation method is one more method
that uses a simple algebraic equation. The advantage of this over the Pearson’s
square method is that it is useful in considering more than two feed ingredients at
a time when balancing rations is more complex. The two by two matrix method
solves two nutrient requirements using two feed ingredients. A 2 × 2 matrix is set
and a series of equations are done to come up with the solution to the problem.
Trial and error method: This method is used for formulating rations for swine
and poultry. As the name implies, the formulation is manipulated until the nutri-
ent requirements of the animal are reached. This method makes possible a diet
formulation that meets all the requirements of the animal. The major limitation
of this method is that it is laborious and time consuming before one approaches
a better solution. Many mathematical programming techniques are being used to
formulate rations, such as linear programming, multiple-objective programming,
goal programming, separable programming, quadratic programming, nonlinear
programming and genetic algorithm. In the present work, we are discussing linear
and nonlinear programming techniques in animal diet formulation.
https://t.me/livestockscience
Review of Literature 11
https://t.me/livestockscience
12 Livestock Ration Formulation
All these assumptions indicate that the objective function and all the constraints
must be characterized by a linear relationship among decision variables in the
LP model. LP has been used as a basic tool for animal feed formulation since
1950. Interest in livestock ration formulation accelerated, and a laboratory diet was
considered as an important variable. During the 1970s, the American Institute of
Nutrition, National Academy of Science, Institute of Laboratory Science, Institute
of Laboratory Animal Resources and Laboratory Animal Centre Diets Advisory
Committee supported the use of standard reference diets in biomedical research.
A linear model was developed (Van de Panne and Popp, 1963) to formulate an
optimized composition of cattle feed. In this LP model, the coeffcients of con-
straints can be considered as stochastic. Feed utilization is described in terms of
feed ration, response gain, biological function and economic effciency. Nutritional,
genetic and physiological responses are interpreted as a function of gain and feed
intake (Meyer and Garrett, 1967); the incorporation of information on animal per-
formance into the linear programming derivation of optimum livestock rations is
taken into consideration (Townsley, 1968). Another model was developed using
linear approximation of chance constrained programming (Olson and Swenseth,
1987). The LP model can be solved for a complicated set of nutrient requirements
to give a relatively well-balanced ration (Haar and Black, 1991: 541–556). Sklan
and Dariel (1993) developed a nutritional program for high producing dairy herds
to attain effcient and proftable levels of milk production. The main goal in the
application of LP in feed formulation is the production of least cost rations that
produce satisfactory results. A nutrition program was developed for high producing
dairy herds to attain effcient and proftable satisfactory results. It has been widely
used in livestock rations (Lara, 1993: 321–334), and to formulate feeds, nutrition-
ists should be knowledgeable in diet specifcations as well as result interpretations.
LP is one of the most important techniques to allocate the available feedstuffs in a
least cost broiler ration formulation (Dantzig, 1951; Aletor, 1986; Ali and Leeson,
1995). Olorunfemi et al. (2001) reviewed extensively the use of linear program-
ming in the least cost formulation for aquaculture. Olorunfemi et al. (2001) also
applied linear programming into duckweed utilization in low-cost feed formula-
tion for broiler starter. In 2002, Darmon used LP as a tool to optimize human
nutrition. A model developed by Tedeschi et al. (2004) is used to represent the
effciency of nutrient use in relation to proftability of dairy farms. Guevara in 2004
developed a cost analysis spreadsheet and validated the effcacy on milking and
custom heifer operations. A stochastic linear programming-based Excel workbook
was developed that consists of two worksheets illustrating linear and stochastic
program methods. This Excel sheet was set up to calculate the margin of safety
value according to probability. The multiple-objective programming was applied
for feed formulation with the objective of minimizing the nutrient and minimizing
cost. This study introduced a dual model in linear programming to obtain the price
of resources that take part in optimization. The price of nutrients resource showed
degree of infuence of a diet’s cost by increasing or decreasing the expected nutrient
values. When the price of some kind of resource is zero, it means that reaching this
nutrient value does not have infuence on a special diet least cost within a particu-
lar value (Xiong et al., 2003). Therefore, for the past seven decades, use of linear
https://t.me/livestockscience
Review of Literature 13
programming technique has been used to formulate least cost dairy feed (Waugh,
1951; Rehman and Romero, 1987; Zhang and Roush, 2002; Hadrich et al., 2008;
Zgajnar and Kavcic, 2008; Al-Deseit, 2009; Almasad et al., 2011; Saxena, 2011a).
https://t.me/livestockscience
14 Livestock Ration Formulation
in practice may not be true. Since varying constraints need to be considered, the
feed mixing problem has become increasingly diffcult to solve using linear pro-
gramming alone. In varying situations, LP is found insuffcient to overcome all the
complexities of the problem. It suffers unnecessarily from over-rigid specifcation
of nutritional and other requirements. Some relaxation of the constraints imposed
would not seriously effect an animal’s well-being and performance.
Over the years, several issues have arisen in feed mixing problems such as ingre-
dient variability. Among them are that the nutrient levels of feed ingredients are
unstable and fuctuate (Munford, 1996; Panne and Popp, 1963), that there can be a
nutrient imbalance in the fnal solution and that price variability means prices of
feed ingredients are not constant, making the solution infeasible. These problems
include the singularity of the objective function and the rigidity of the constraint
set. The singularity of the objective function refers to the reliance on cost alone as
the most important factor in determining the composition of the ration. Lara (1993)
also criticizes practical applications of LP due to the restrictions placed on decision
makers’ preferences through a singular objective function. Producers are likely to
have many objectives in mind while formulating a ration (Tozer and Stokes, 2001).
Mitani and Nakayama (1997: 131–139) pointed out three limitations of linear pro-
gramming model while formulating the ration:
https://t.me/livestockscience
Review of Literature 15
and/or some or all of the constraints are nonlinear functions. Such problems are
referred as nonlinear programming problems (NLPP). If the objective function and
the constraints are all convex functions (a function which, when plotted, results in
a curve always curving upwards or not curving at all is called a convex function),
the problem is said to be a convex programming problem. The method of solving
an LP problem is based on the property that the optimal solution lies at one or more
extreme points of the feasible region. This limits our search to corner points, and
a fnite solution is obtained after a fnite number of iterations, as in the simplex
method. But this is not true for an NLPP. In such problems, the optimal solution
can be located at any point along the boundaries of the feasible region or even
within the region. And due to nonlinearity of the objective function and constraints,
it becomes diffcult to differentiate between the local and global solution. Every
linear programming problem can be solved by the simplex method, but there is no
single technique which can be claimed to effciently solve each and every nonlinear
optimization problem. In fact, a technique which is effcient for one nonlinear opti-
mization problem may be highly ineffcient for solving another NLPP. A variety
of computational techniques for solving NLPP are available (Himmelblau, 1972;
McCormick, 1970). However, an effcient method for the solution of a general NLPP
is still a subject of research. There is a wide variety of nonlinear programming
problems. Some of the most important types are constrained and unconstrained
NLPP, separable programming, convex programming, non-convex programming,
quadratic programming and so on. In a linear-constrained NLPP, all the constraints
are of a linear type, and the objective function is nonlinear. And in a nonlinear-
constrained NLPP, all the constraints are nonlinear. Many have tried to use uncon-
strained optimization methods for solving nonlinear problems with constraints. A
successful and frequently used approach is to defne an auxiliary unconstrained
problem such that the solution of the unconstrained problem yields the solution
of the constrained problem. Goldfarb has extended the Davidon–Fletcher–Powell
method to handle problems with linear constraints utilizing the concept of gradi-
ent projection (Goldfarb, 1969: 739–764). The method was generalized by Davies
to handle nonlinear constraints (Davies, 1970). Box developed a constrained ver-
sion of the simplex method (Broyden, 1967: 386–381). Another method that uses
the simplex technique in constrained optimization was proposed by Dixon (1973a:
23–32). Another important class of methods used for solving a constrained NLPP
is known as penalty function methods. In these methods the constrained problem
is converted into an unconstrained problem or a sequence of unconstrained prob-
lems in which there is a severe penalty for the violation of the constraints. Another
class of methods for solving an NLPP is the family of exact penalty function meth-
ods. The pioneering work in this feld has been done by McCormick (1983). The
sequential linear programming approach for constrained optimization problems
fnds the optimal solution by repeated linear approximation of a nonlinear prob-
lem and using linear programming techniques to solve it. Attempts have also been
made to obtain the solution of both constrained and unconstrained optimization
problems by solving the Kuhn–Tucker (KT) conditions directly. Methods under this
category are sometimes classifed as multiplier methods. Quasi-Newton type meth-
ods have been used to obtain the solution to the system of equations representing
KT conditions (Powell, 1977). He also tried to obtain solutions to constrained and
https://t.me/livestockscience
16 Livestock Ration Formulation
unconstrained NLPP by using a least square algorithm for solving the system of
nonlinear equations representing the KT condition of the problem. In the absence
of convexity, the methods discussed here at the most guarantee a local optimal
solution. Keeping in view the practical necessity and the availability of fast com-
puting machines, many computational techniques are being developed to fnd the
global optimal solution. The methods for solving global optimization problems are
broadly classifed as deterministic and probabilistic methods. The deterministic
methods try to guarantee that a neighborhood of the global optima is attained. Such
methods do not use any stochastic techniques but rely on a thorough search of the
feasible domain. They are applicable to a restricted class of functions only, such as
Lipschitz continuous functions. In stochastic or probabilistic methods, two phases
are generally employed. In the frst phase, or the global phase, the function is evalu-
ated at a number of randomly sampled points. In the second phase, also called the
local phase, these points are manipulated by local searches to yield a possible can-
didate for the global minima. These methods are preferred over the deterministic
methods because they are applicable to a wider class of functions. Some of the
earlier methods of this category are pure random search methods by Bremermann
and Anderson (1989) and Brooks (1958: 244–251). Dixon suggested a method in
which any local search method can be used to search the global solution by start-
ing repeatedly from different initial points chosen stochastically. Price presented a
controlled random search technique (RST) method in which the simplex approach
is used on a random sample of points to yield a better point at each iteration and the
method in the limit converges to the point of global minima Dixon (1972a, 1972b,
1973a, 1973b, 1976) and Bharti (1994) have updated the RST method of Price for
solving the constrained NLPP (Dixon 1973a).
Bharti has tried to improve the performance of various controlled random
search algorithms (1994). Linear programming is an effective method of fnding
the least possible cost of a unit that satisfes a set of nutrient specifcations. An
animal’s nutritional requirements hold constant only within a single performance
level. Changes in diet composition may cause variations in feed intake, animal’s
nutritional requirements and animal performance. The traditional LP formulation
ignores animal performance and feed intake. There are multiple least cost diets at
different levels of nutrient concentrations and corresponding animal performances.
Thus, a single LP formulation will not necessarily converge to the minimum cost
of production or maximum proft, both nonlinear functions of diet composition
(Tedeschi et al., 2005). Low-cost balanced ration formulation for livestock involves
omitting ingredients with an inclusion rate below some fxed threshold and round-
ing other ingredients to realistic weighing quantities. It is possible to incorporate
these constraints, but then it will not be possible to solve by linear programming,
as the nonlinear effect of variables will then be included (Saxena, 2011c: 1–5). It
is necessary to control a ratio of nutrients in a formulation – for example, a simple
ratio such as that of calcium to phosphorus, or a more complicated one such as
forage dry matter to concentrated dry matter in a ruminant complete diet. Also
controlling the dry matter percentage as-fed in a diet involving wet feeds is ratio
constraint. Ratio constraints are nonlinear and fall outside the scope of a linear
programming framework (Munford, 2005, 1989).
https://t.me/livestockscience
Review of Literature 17
https://t.me/livestockscience
18 Livestock Ration Formulation
is used to solve two NLPPs of animal diet formulation (Tozer, 2000: 443–451). To
overcome the drawback of linear approximation of the objective function for diet
formulation, a mathematical model based on the NLP technique was developed
by Pratiksha Saxena to measure animal performance in terms of milk yield and
weight gain. The result of developed model was compared with the linear model,
and the comparison shows that NLP gives better results for maximization of animal
yield and weight gain and represents simultaneous effects of all variables together
(Saxena, 2011a: 106–108).
Here di+ is the positive deviation variable from overachieving the ith goal; di− is the
negative deviation variable from underachieving the ith goal, and xj is the decision
variables; aij is the decision variable coeffcient. In the weights method, the single
objective function is the weighted sum of the functions representing the goals of the
problem. The weights goal-programming model is of the form:
where di+, di−, xj ≥ 0 and (i = 1,2. . .m), ( j = 1,2. . .n). Here Wi+ and Wi− are non-
negative constraints and can be real numbers representing the relative weights
assigned within a priority level to the deviational variables. The parameter di+ rep-
resents positive weights that refect the decision maker’s preference regarding the
relative importance of each goal, while di− is the negative weights of the decision
https://t.me/livestockscience
Review of Literature 19
maker’s preference. The determinant of the specifc values of these weights is sub-
jective. The objective of goal programming is to minimize the deviation to fnd the
solution for which the deviation is at minimum. Therefore, to overcome the limita-
tion of LP in diet formulation, a multiple criteria decision-making technique was
introduced by Romero and Rehman (1984). In 1992 Lara and Romero used the con-
cept of Rehman and Romero (1987) and then applied a multiple goal programming
(MGP) model to introduce the relaxation of nutrient constraints. Lara and Romero
(1994) extended their work, as the relaxation of constraints could reduce the ration
cost. They considered nutritional requirements as targets which might or might
not be achieved. This MGP model is solved using a multiple criteria mathematical
programming technique known as the interactive method. The use of this interac-
tive method is to perform a search over a set of feasible diets. The work of Mitani
and Nakayama (1997) is also an early contribution in diet formulation. Lara (1993)
introduced a second objective, maximization of the inclusion in the diet of the
ingredients available on the farm in addition to least cost. This work is introduced
in a nonlinear framework because the second objective is considered in fractional
form. Tozer and Stokes (2001) used multiple objective programming to reduce
nutrient excretion from dairy cows through incorporation of a nutrient excretion
function into the ration formulation framework. To reduce the nutrient excretion
load, rations are formulated to minimize cost and nitrogen and phosphorus excre-
tion using MGP. Jean dit Bailleul et al. (2001) also developed a multiple objective
optimization method to minimize the cost and excess nitrogen in pig diets. Zhang
and Roush (2002) introduced two objectives minimizing the cost and the nutrient
variabilities for protein, methanone and lysine. MGP applies the same concepts as
GP but is different in terms of modelling the objective separately. Therefore, GP
and MGP have an advantage in handling multiple objectives, including nutrient
variability and reducing nutrient imbalance problems. Castrodeaza et al. (2005)
formulated pig rations considering economic and environmental objectives, incor-
porating advanced nutritional concepts in ratio form and by using an interactive
method by multiple objective fractional programming. They considered the cost
of feed, the lysine/energy ratio, deviation with regard to the ideal values of the
percentage content of amino acids in the protein and the amount of phosphorus.
Zgajnar and associates (Zgajnar and Kavcic, 2008, 2009; Zgajnar et al., 2009) also
combined LP and weighted GP to fnd the optimal ration cost with balanced nutri-
tional requirement. One of the primary objectives was to overcome the limitation
of linear programming such as LP rigidity and not satisfying primary constraints.
https://t.me/livestockscience
20 Livestock Ration Formulation
during the process, which is called selection. The new chromosomes are created
using genetic operators such as crossover and mutation [Goldberg (1989), Holland
(1975)]. GAs has great history of success in search problems; the reason for this is
their ability to exploit the information about an initially unknown search space in
order to bias subsequent searches into useful subspaces, that is, their adaptation.
This is the key factor, particularly in large, poorly understood search space prob-
lems where classical and old conventional techniques are inappropriate in offering
a good search space. Although there are many variants of GA, the fundamental
mechanism operates on a population of individuals and consists of three opera-
tions: a) creating initial population, b) evaluation of individual ftness, c) formation
of a gene pool by selection procedure and d) recombining them using crossover and
mutation procedures. In 1998, F. Herrera et al., reviewed the features of real-coded
genetic algorithms in which different models of genetic operators and some mecha-
nisms available for studying the behavior of this type of GA have been compared.
This review stated that initially the use of RCGA appeared in applications such as
chemo metric applications (Lucasius and Kateman, 1989) and use of meta operators
(Davis, 1989). Real coded GA has been used mainly for numerical optimization in
continuous domains. Later, in 1994, Darrell Whitey covered the canonical genetic
algorithm as well as more experimental forms of genetic algorithm, including
parallel island models and parallel cellular genetic algorithms. He also illustrates
genetic searches by hyperplane sampling and reviewed the theoretical foundation
of genetic algorithms which include the fundamental theorem of a genetic algo-
rithms called a “schema theorem” (Holland, 1975) as well as an exact model of the
canonical genetic algorithm. Therefore, GA has been used for solving nonlinear
programming models as well as multi-objective programming models. It focuses
on the initialization process, evaluation process, selection process, crossover and
mutation for the nonlinear goal programming problem and concluded that GA is
effective for nonlinear GPPs (Zheng et al., 1996; Gen and Cheng, 1997; Deb, 1999;
Kumar et al., 2012). In 2014 a combined LP with GP for feed formulation displayed
the dip in cost as a major advantage of using the goal-programming approach
(Ghosh et al., 2014). Therefore, in view of an exhaustive review of literature, it
is observed that no techniques are suitable for formulating rations in a least cost
manner for dairy cattle; thus, a comparative study was planned to fnd a suitable
technique that is farmer friendly and can be adopted at the farm level.
https://t.me/livestockscience
3
Tools and Techniques
CONTENTS
3.1 Computation of LP Model Using MS-Excel Based Solver .......................... 21
3.1.1 Simplex LP Method in MS-Excel Solver ........................................ 22
3.1.2 GRG Nonlinear in MS-Excel Solver ............................................... 22
3.1.3 Evolutionary Algorithm in MS-Excel Solver .................................. 22
3.1.4 MATLAB (Matrix-Laboratory) ...................................................... 23
3.1.4.1 Inputs for Genetic Algorithm........................................... 24
3.2 Data Handling and Nutrient Bound Calculator ........................................... 26
DOI: 10.1201/9781003231714-3 21
https://t.me/livestockscience
22 Livestock Ration Formulation
https://t.me/livestockscience
Tools and Techniques 23
https://t.me/livestockscience
24 Livestock Ration Formulation
population where, in every step, it randomly selects individuals from the present
population and uses them as a parent to create new children for the next generation.
After many generations the set of the population tends toward an optimal solution.
There are many options and inbuilt functions in MATLAB to modify the param-
eters of the genetic algorithm according to the nature of the optimization problem.
Genetic algorithm differs from old classical methods because old classical methods
create only one point at every iterative step where a string of points approach the
best-ft solution, whereas genetic algorithm creates a population at every iteration,
and the best-ft individual approaches the optimal solution. In MATLAB using an
inbuilt function, in general, genetic algorithm can be defned as follows:
Gaoptimset provides the options for GA parameters. Populace type represents the
type of data in the population, and by default it takes “double vector”. Creating an
initial population is very important for the algorithm, as the entire population set is
https://t.me/livestockscience
Tools and Techniques 25
based on initial population; hence, the creation function creates the initial popula-
tion in which “gacreationlinearfeasible” is the best option to create the initial popu-
lation which satisfes all bounds as well as linear constraints. Therefore, we have
created an initial population in the range of lower and upper bounds. The popula-
tion size describes how many individuals should be there in each generation. Even
though giving a large population size makes the algorithm run slowly, it is better to
keep a large population size, as by keeping a large population size we are allowing
an algorithm to search in a broader space and there is less chance for the algo-
rithm to stop at local optima. Elite count describes the count of individuals that are
assured to remain in every genesis; it takes only positive integers, or if the value is
not assigned, it takes the default value as a ceiling (0.05 × population size). Elitism
is very important in respect to convergence, as by allowing at least two or three
individuals to survive at every generation, it makes the algorithm converge faster.
Crossover is a very important parameter of the genetic algorithm as it combines two
best-ft parents to create new offspring at every generation. “CrossoverFcn” options
defne the crossover type. There exist various kinds of crossover, such as 1-point
crossover, 2-point crossover etc., but for linear constraints and bounds, “crossover-
heuristic” is the best option for creating offspring. “Crossoverheuristic” returns
offspring which lie on the line that involves two parents; one can specify how far
the children is from the better parent by the parameter ratio. MATLAB takes the
default value of this ratio as 1:2. In general, let P1 and P2 be the two parents and
P1 has high quality robust value, crossoverheuristic will return the offspring by
the formula P2 + ratio × (P1 − P2). The crossover fraction gives the division of each
populace children that are other than world class children (elite children) which
are comprised of crossover children. If the value of the crossover fraction is 1, it
reveals that the algorithm is not having mutation, whereas if the crossover frac-
tion is 0 it refects that algorithm is running without crossover. Therefore, it is a
good choice to keep the value of crossover fraction in between 0.6–0.8 for better
diversity. Like crossover, mutation is also an important parameter of genetic algo-
rithm to maintain the diversity of population; it allows the algorithm to search in a
broader space. Mutation generally help the algorithm make small adjustments in an
entity to create a mutation child. One can defne the mutation option by the function
name “MutationFcn”. Due to rigidity in linear constraints, preference is given to
“mutationadaptfeasible”, where an adaptive feasible mutation generates a direction
which is fexible compared to a previous lucrative or doomed genesis. This muta-
tion chooses a direction and step length that satisfes both linear constraints as well
as bounds. The selection operator helps the algorithm to select best-ft parents for
the next generation; one can specify this option with the function name “selection-
Fcn”. There are many types of selection procedures, such as roulette wheel selec-
tion, uniform selection etc., but preference is given to tournament selection, as the
algorithm will choose each parent by its tournament size 2 at random, and then the
best individual out of these will be selected as parent. Once these parameters are
defned, the genetic algorithm Solver is called to run the program. These param-
eters can be tested on different types of optimization problems (linear or nonlin-
ear), but these parameters need to be modifed as per the requirements. In Chapters
5 and 6, the performance of this MATLAB-based genetic algorithm is tested on
https://t.me/livestockscience
26 Livestock Ration Formulation
linear and nonlinear goal programming problems for least cost ration formulation
(Kuntal et al., 2016).
https://t.me/livestockscience
4
Binary Coded Genetic Algorithm to
Solve Ration Formulation Problem
CONTENTS
4.1 Introduction .................................................................................................. 27
4.2 Linear and Nonlinear Programming Model for Animal Feed
Formulation (Test-Problem-1) ...................................................................... 28
4.2.1 Linear Programming Formulation .................................................. 28
4.2.2 Nonlinear Programming Formulation............................................. 29
4.2.3 Extended Work Using Genetic Algorithms ..................................... 29
4.2.4 Solution of Nonlinear Programming Model by Genetic
Algorithm ........................................................................................ 30
4.2.5 Graphical Results ............................................................................ 32
4.3 Goal-Programming Model (Test Problem-2) ............................................... 33
4.3.1 Formulation of the Nonlinear Goal Programming Problem ........... 33
4.3.2 Graphical Results ............................................................................ 39
4.3.3 Conclusion ....................................................................................... 42
4.1 Introduction
The farmers of India hold a small number of dairy animals, in which cattle and
buffalo are the main animals used for milk production. With the increase in the
number of high milk producing dairy animals in different part of India due to many
cross-breeding programs, there is a shortage of feed ingredients for dairy animals,
due to which farmers are more conscious about scientifc animal feeding prac-
tices. In addition to that, because of environmental hazards and global warming
problems due to animal excreta, balancing nutrition production for proper health
and excretion is required. Many qualifed animal nutritionists are also fnding dif-
fculty in formulating the low-cost balanced ration for growing dairy animals at
different body conditions. Even though linear programming is a good method for
formulating least cost diet, it has many limitations, as the linear programming tech-
nique requires the objective function to be single and constraints to be rigid (RHS).
Recently, many researchers have described the benefts of the goal programming
method for ration formulation in which the weighted sum goal programming with
a penalty function is used to solve the problem. In this research, an attempt is made
to test the effectiveness of a heuristic algorithm such as a genetic algorithm to opti-
mize the ration formulation on nonlinear and goal programming models.
DOI: 10.1201/9781003231714-4 27
https://t.me/livestockscience
28 Livestock Ration Formulation
TABLE 4.1
Composition of Concentrate Mixtures in Respect of DCP and TDN
Control group Cotton seed Cotton seed cake Cotton seed cake
Ingredients (groundnut cake) (whole) (undecorticated) (decorticated)
Groundnut cake 22 10 0 0
Cotton seed 0 57 0 0
(undecorticated) 0 0 44 0
(decorticated) 0 0 0 27
Wheat bran 75 30 53 70
Common salt 2 2 2 2
Mineral mixture 1 1 1 1
https://t.me/livestockscience
Binary Coded Genetic Algorithm 29
This LP model is solved by the simplex method, which leads to the solution as x1 =
782.97800, x2 = 75.943, x3 = 508.9343 gm/kg metabolic body weight, with maxi-
mum objective value of 24.55.
This nonlinear model is solved by using the Kuhn-Tucker method, which gives
the values of the three nutrient ingredients as x1 = (782.97800), x2 = (67.00717),
x3 = (507.79209) gm/kg metabolic body weight, with a maximum objective value of
582.01 gm/kg metabolic weight.
https://t.me/livestockscience
30 Livestock Ration Formulation
points, and in the local phase, these feasible points are controlled by local search
options to produce a best ft individual for global minima. GA is an optimization
method which is used to optimize both constrained and unconstrained optimization
problems; it is an adaptive heuristic algorithm which works on Darwin’s principle of
survival of the fttest. The idea behind genetic algorithm is to simulate the process in
the natural system which is necessary for evolution. GA not only provides an alterna-
tive method but also outperforms the other traditional methods consistently. GA starts
with set of solutions (chromosomes) called the population. Solution from one set of
population has been taken to create a new set of population with the hope that the
new set of population is better than the old one. This new set of solutions (offspring)
will be chosen from the population based on its best ftness value. The robust value
is allocated to every individual, where computation of robust value depends upon the
application. At every iteration (generations), candidates are chosen from the set of
randomly generated population for reproduction of new offspring by crossover and
mutation, where crossover will be done with high probability compared to mutation.
The selection of individuals will be depending upon high ftness, but the mean ftness
of population likes to improve the solution from generation to generation. Due to its
random nature, GA improves the chance of fnding the global optimum solution. We
have slightly modifed the technique to suit our requirements and solved the nonlinear
programming problem described by Pratiksha Sexena.
TABLE 4.2
Solution for NLP Model by GA for Original Bounds (Test Problem-1)
x1 x2 x3
(gm/kg metabolic (gm/kg metabolic (gm/kg metabolic Objective
Iterations weight) weight) weight) function
https://t.me/livestockscience
Binary Coded Genetic Algorithm 31
x1 x2 x3
(gm/kg metabolic (gm/kg metabolic (gm/kg metabolic Objective
Iterations weight) weight) weight) function
TABLE 4.3
Solution for NLP Model by GA for Reduced Bounds (Test Problem-2)
x1 x2 x3
(gm/kg metabolic (gm/kg metabolic (gm/kg metabolic Objective
Iterations weight) weight) weight) function
TABLE 4.4
Comparison of Solution Obtained by GA and RST for NLP Model
RST2 GA
https://t.me/livestockscience
32 Livestock Ration Formulation
760
740
720
700
680
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
No of Iteraions
FIGURE 4.1 Maximum Milk Yield Using GA (Original Bounds); Kuntal, IJESET, 2003.
590
588
586
584
582
580
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
No of Iteraions
FIGURE 4.2 Maximum Milk Yield Using GA (Reduced Bounds), Kuntal, IJESET, 2003
Graphical Representation of Maximum Milk Yield Using RST and GA (original Bounds)
950 GA
900 X: 3000 RST
Objective function
850 Y: 819.2
800
750
700
650 X: 700
600 Y: 569.9
550
500
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
No of Iteraions
FIGURE 4.3 Maximum Milk Yield Using RST and GA (Original Bounds), Kuntal, IJESET, 2003
https://t.me/livestockscience
Binary Coded Genetic Algorithm 33
610 X: 500
600 Y: 593.8
590
580 X: 700
570 Y: 566
560
550
540
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
No of Iteraions
FIGURE 4.4 Maximum Milk Yield Using RST and GA (Reduced Bounds), Kuntal, IJESET, 2003
https://t.me/livestockscience
34 Livestock Ration Formulation
assigned to every goal. In this nonlinear GP model with nonlinear objective function
(Gupta et al., 2013b), goals are allotted in such a fashion that all +ve deviational vari-
ables (≤), −ve deviational variables (≥) and both +ve and −ve deviational variables (=)
need to be minimized. The brief description of the goals requirements is as follows:
G-1: dlc−, dlc+ represent an under- and overachievement of least cost ration,
where dlc+ requires minimization.
G-2 and 3: dt−, dt+ represent under- and overachievement of total weight of
feed ingredients, where both require minimization.
G-4: dcp− and dcp+ represent under- and overachievement of the protein
requirement, where dcp− requires minimization.
G-5: dtdn− and dtdn+ represent under- and overachievement of the energy
requirement, in which dtdn− requires minimization.
G-6: dca− and dca+ represent under- and overachievement of the calcium
requirement, where dca− requires minimization.
G-7: dp− and dp+ represent under- and overachievement of the phosphorus
requirement, where dp− requires minimization.
G-8: dg− and dg+ represent under- and overachievement of the grain, maize
and jowar requirement, where dg+ requires minimization.
G-9: db− and db+ represent under- and overachievement of the bran rice and
wheat requirement, in which db+ requires minimization.
G-10: dck− and dck+ represent under- and overachievement of the groundnut cake
as well as the cotton cake requirement, where dck+ requires minimization.
G-11 and 12: dr/c− and dr/c+ give under- and over-achievement of roughage as
well as concentrate ratio (that is, 2:3), where both the deviations require
minimization.
G-13 and 14: dd/g− and dd/g+ represent under- and overachievement of dry-
green ratio (that is, 2:3), where both the deviations need to be minimized.
G-15 and 16: dc/h− and dc/h+ represent the under- and overachievement ratio
of legume and non-legume (that is, cowpea: hybrid Napier in ratio of 1:1),
where both the deviations require minimization. The GP model is nonlinear,
as the objective function involves the square root of the sum of the squares of
the deviations, in which the weight is allocated to every goal as per the prior-
ity (Gupta et al., 2013b). The goal programming model has 12 goals and nine
decision variables, with 24 deviational variables. The objective function has
been formulated by considering deviational variables for each goal (maxi-
mization or minimization) whose target is given on RHS. The nonlinear GP
model for all three levels of animals is given in Table 4.5 where,
https://t.me/livestockscience
TABLE 4.5
Min(z) = P1d l c +2 + P2 d t −2 + P3d t +2 + P4 d cp +2 + P5d tdn−2 + P6 d ca−2 + P7 d p−2 + P8 d g +2 + P9 d b +2 + P10 d ck +2 + P11d r −2 + P12 d r +2 + P13d d −2 + P14 d d +2 + P15d c −2 + P16 d c +2
c c g g h h
=1 =1 =1
Total (kg): x1 + x2 + x3 + x4 + x5 + x6 + x7 + x8 + x9 + dt − − dt +
Protein (g/kg): ≥31 ≥108 ≥46.53
30x1 +10x2 +180x3 + 80x4 +110x5 +120 x6 +120x7 + 450x8 + 300x9 + dCP − − dCPP +
Grain max @ 70% of concentrate (kg): x4 + x5 + dg− − dg+ ≤0.36 ≤0.36 ≤0.36
Bran max @ 50% of concentrate (kg): x6 + x7 + db− − db+ ≤0.30 ≤0.30 ≤0.30
Cake max @ 30% of concentrate (kg): x8 + x9 + dCK− − dCK+ ≤0.21 ≤0.21 ≤0.21
Roughage /concentrate: 3(x1 + x2 + x3 ) − 2( x4 + x5 + x6 + x7 + x8 ) + d r − + =0 =0 =0
− dr
c c
Dry/green roughages: 3x1 − 2(x2 + x3 ) + d d − + =0 =0 =0
− dd
g g
=0 =0 =0
35
− +
Legume/non-legume greens: x2 − x3 + d c − dc
h h
https://t.me/livestockscience
TABLE 4.6
Solution Set of LPP for Level-1, 2 and 3 by GA
36
Generation Levels x1 x2 x3 x4 x5 x6 x7 x8 x9 Z
100 L-1 0.1174 0.0465 0.1002 0 0.2384 0 0.2627 0 0 0.2784
L-2 0.0819 0.0638 0.0813 0 0.3225 0 0.0278 0 0.1 0.0134
L-3 0.0649 0.0396 0.0422 0 0.3142 0 0.2633 0 0 0.5578
200 L-1 0.0678 0.0429 0.0718 0 0.3336 0 0.2201 0 0 0.2861
L-2 0.064 0.0367 0.0959 0 0.3338 0 0.0155 0 0.0623 0.0149
L-3 0.093 0.0381 0.0995 0 0.2960 0 0.2378 0 0.0000 0.586
300 L-1 0.0664 0.0289 0.1258 0 0.3253 0 0.2831 0 0 0.2814
L-2 0.0599 0.0394 0.0516 0 0.3175 0 0.0228 0 0.1155 0.0159
L-3 0.0775 0.0518 0.0857 0 0.2855 0 0.2477 0 0 0.6303
400 L-1 0.0756 0.0327 0.0741 0 0.2387 0 0.2286 0 0 0.2656
L-2 0.0834 0.0509 0.0648 0 0.3389 0 0.0312 0 0.0539 0.0102
L-3 0.0930 0.0381 0.0995 0 0.2960 0 0.2378 0 0 0.5860
500 L-1 0.1039 0.0311 0.0758 0 0.2815 0 0.2605 0 0 0.2852
L-2 0.0534 0.0538 0.1118 0 0.3261 0 0.0194 0 0.0849 0.0136
L-3 0.0378 0.0299 0.1477 0 0.2983 0 0.2648 0 0 0.6098
600 L-1 0.1027 0.0362 0.0943 0 0.3287 0 0.2692 0 0 0.2767
L-2 0.1106 0.0547 0.0655 0 0.3572 0 0.0269 0 0.0957 0.0113
https://t.me/livestockscience
TABLE 4.7
Solution Set of the Nonlinear Programming Problem (NLPP) for Level-1, 2 and 3 by GA
37
L-2 0.0531 0.0553 0.0805 0 0.254 0 0.025 0 0.041 0.027
L-3 0.0903 0.0328 0.0863 0 0.289 0 0.273 0 0 0.552
https://t.me/livestockscience
38
TABLE 4.8
Comparison of Decision Variables by RST and GA
GA RST
https://t.me/livestockscience
Binary Coded Genetic Algorithm 39
Goal 5
0 Goal 6
Goal 7
Goal 8
-0.5
Goal 9
Goal 10
100 200 300 400 500 600 700 800 900 1000 Goal 11
Generations Goal 12
Goal 13
Goal 14
Goal 15
Goal 16
0.2
Goal 5
0 Goal 6
Goal 7
-0.2 Goal 8
Goal 9
200 400 600 800 1000 Goal 10
Generations Goal 11
Goal 12
Goal 13
Goal 14
Goal 15
Goal 16
0.4
Goal 5
0.2 Goal 6
Goal 7
0 Goal 8
-0.2 Goal 9
200 400 600 800 1000 Goal 10
Generations Goal 11
Goal 12
Goal 13
Goal 14
Goal 15
Goal 16
FIGURE 4.5 Graphs of Deviations for Linear Model of Level-1, 2 and 3, Kuntal, IJEIT, 2013.
https://t.me/livestockscience
40 Livestock Ration Formulation
Goal 5
0.1 Goal 6
0 Goal 7
-0.1 Goal 8
Goal 9
-0.2 Goal 10
Goal 11
100 200 300 400 500 600 700 800 900 1000 Goal 12
Generations Goal 13
Goal 14
Goal 15
Goal 16
-3
x 10 Non-linear Programming graph of level 2 Goal 1
1.5
Goal 2
1 Goal 3
Goal 4
0.5
Deviations
Goal 5
Goal 6
0
Goal 7
-0.5 Goal 8
Goal 9
-1
Goal 10
-1.5 Goal 11
100 200 300 400 500 600 700 800 900 1000 Goal 12
Generations Goal 13
Goal 14
Goal 15
Goal 16
0.2 Goal 5
Goal 6
0 Goal 7
Goal 8
-0.2
Goal 9
-0.4 Goal 10
Goal 11
100 200 300 400 500 600 700 800 900 1000 Goal 12
Generations Goal 13
Goal 14
Goal 15
Goal 16
FIGURE 4.6 Deviations for Nonlinear GP Model of Level-1, 2 and 3, Kuntal, IJEIT, 2013
https://t.me/livestockscience
Binary Coded Genetic Algorithm 41
0.4 GA
RST
Deviations
0.2
1 2 3 4 5 6 7 8 9
No of Variables
0.4 GA
RST
Deviations
0.2
1 2 3 4 5 6 7 8 9
No of Variables
0.6 GA
RST
Deviations
0.4
0.2
1 2 3 4 5 6 7 8 9
No of Variables
https://t.me/livestockscience
42 Livestock Ration Formulation
4.3.3 Conclusion
Solutions of NLPP (1) and (2) with original and reduced bounds using RST reveals
that the variation in the value of decision variables x1, x2 and x3 are 4.3%, −0.19%
and 20.17% respectively, which leads to 0.68% variation in the objective function
(maximum milk yield). There is an approximately 2% deviation in the objective
function in comparison with results obtained by Pratiksha Sexena using the Kuhn-
Tucker condition.
Solutions of NLPP (1) and (2) using the genetic algorithm (see Tables 4.2 and
4.3) reveals that the variation of decision variables x1, x2 and x3 are 9.3%, 6.9% and
6.08%, respectively, which leads to 27% variation in the objective function, which
is more compared to RST2. Hence, a reduction of bounds gives a better result,
which is comparable with RST2 as there is a deviation of 2% (approximately) in
the objective function with the value obtained by the Kuhn-Tucker method. Using
a wide range of bounds for GA gives better results only after 3000 generations,
whereas after reducing the bounds, the best result is obtained in 500 generations.
After comparing the results obtained by GA with RST2, it reveals that both applied
techniques (RST2 and GA) are nearly the same, with a small difference of 4.6%
(See Figures 4.1, 4.2, 4.3 and 4.4). The possible range of choice variables and goal
characteristics is given in Table 4.4. Therefore, there is a chance of solving the non-
linear model of animal feed formulation using GA with a slight modifcation in the
technique according to our requirements. GA gives adaptability to pick any pos-
sible solutions from the vast domain of the solution set, which adds an extra bonus
for the planner. In this study, the NLP model is solved by using a genetic algo-
rithm for 1000 generations for each level of dairy animals. The result is shown in
Tables 4.6 and 4.7 for the LP and NLP models. Moreover, the comparison of the
controlled random search technique (RST) and the genetic algorithm (GA) for the
nonlinear goal programming problem of each level is shown in Table 4.8. This
investigation centers around a heuristic approach for improvement of the ration
of Indian dairy cows with 10 litres of daily milk yield, considering the nonlinear
weighted sum goal programming formulation. The GA approach is refected as
a better approach in solving nonlinear goal programming problems in compari-
son with the prior methodologies. In the prior work by Shrabani (Shrabani Ghosh
[M.Phil. dissertation] 2011) the linear goal programming problem was solved using
Excel Solver. An endeavour has been made in the present investigation to expand
the linear GPP as a nonlinear weighted sum goal programming problem with pre-
defned priorities at different levels and is solved using the GA approach. Jyothi
Lakshmi likewise made an endeavour to tackle this issue with the controlled ran-
dom search technique (RST). The comparison demonstrates that the GA approach
gives a better outcome as compared to RST (see Figure 4.7). It is likewise seen from
the outcomes appeared in Table 4.6 that Goal-1 is overachieved by 46% to 50% of
variation and Goal-15 is overachieved by 5% to 8.5% of variation for Level-1. In
Level-2, Goal-16 is overachieved by 3.2% to 4.7% of variation. Also, in Level-3
Goal-1 is overachieved by 42.6% to 50.9% of variation. All other goals are achieved
in a thousand generations using GA for the linear model. Figure 4.5 can be alluded
to to see the wholistic perspective of the goal achievements. From Figure 4.6 we
https://t.me/livestockscience
Binary Coded Genetic Algorithm 43
can observe that Goal-1 is overachieved by 16.2% to 20.2% and Goal-15 is over-
achieved by 14.9% to 21.3% of variation for Level-1. For Level-2, all the goals are
achieved, and for Level-3, Goal-1 is overachieved by 46% to 51.8% and Goal-16
is overachieved by 5% to 10% of variation. All other goals are almost achieved in
1000 generations using GA for the NLP model. Thus, by considering these out-
comes for the LP model, our tool gives more profcient rations by tweaking the
nutritive goals and by taking innocuous deviations into consideration by treating
them as underachievement and overachievement. The connected approach of the
nonlinear weighted goal programming problem ends up being a helpful “motor” to
formulate the least cost ration without any nutritive defciency, which is the basic
downside of the LP. The profcient ration can be improved further by adjusting the
target values of the goals, but this tool may be useful for the assessment of conse-
quences due to globalization impacts (input price rise, price volatility and environ-
mental as well as climate change aspects).
e
nc
ie
k sc
oc
st
li ve
e/
t.m
s ://
tp
ht
https://t.me/livestockscience
https://t.me/livestockscience
5
Least Cost Feed Formulation for Dairy
Cattle during Pregnancy by Using
Real Coded Genetic Algorithm
An Application
CONTENTS
5.1 Introduction.................................................................................................. 45
5.1.1 Data Collection and Methodology .................................................. 46
5.1.1.1 LP Model for Cattle-1 ...................................................... 50
5.1.1.2 LP Model for Cattle-2 ...................................................... 50
5.1.1.3 LP Model for Cattle-3 ...................................................... 51
5.2 Implementation ............................................................................................ 52
5.2.1 Results and Discussions................................................................... 53
5.2.2 Conclusion ....................................................................................... 59
5.3 A Goal Programming Approach to Ration Formulation Problem for
Indian Dairy Cows ....................................................................................... 60
5.3.1 Goal Programming Approach ......................................................... 62
5.3.1.1 GP Model-1 ...................................................................... 63
5.3.1.2 GP Model-2 ...................................................................... 63
5.3.1.3 GP Model-3 ...................................................................... 64
5.3.2 Real Coded Genetic Algorithm with Hybrid Function ................... 65
5.3.3 Results ............................................................................................. 66
5.3.4 Discussion........................................................................................ 67
5.3.5 Conclusion ....................................................................................... 68
5.1 Introduction
In the Mandya district of Karnataka, dairy farmers have limited feedstuffs.
Therefore, to fulfll nutrient requirements in terms of crude protein (CP), total
digestible nutrients (TDN), calcium (Ca) and phosphorus (P), it was necessary
to adopt a software program. There are many techniques/software that were
adopted to fulfll the nutrient requirements of dairy animals (www.scialert.net);
however, there is no specifc technique suitable for formulating least cost rations
for dairy animals. Hence, a comparative study was planned to fnd a suitable tech-
nique which is farmer friendly and can be used at the dairy farm level. The focus
of this chapter is to provide a least cost diet formulation based on the nutrient
DOI: 10.1201/9781003231714-5 45
https://t.me/livestockscience
46 Livestock Ration Formulation
requirements of cattle which has been calculated according to the Indian Council
of Agricultural Research (ICAR-NIANP, 2013) and National Research Council
(NRC, 2001) standards during the third trimester of pregnancy. A linear program-
ming model (LP) was formulated for three different types of cattle with body
weight of 500 kg each and the requirement of a 10-litre milk yield with 4% of
fat content, based on the seventh, eighth and ninth months of pregnancy on a dry
matter basis. The LP model has been solved by a general reduced gradient (GRG)
nonlinear, an evolutionary algorithm (EA), a simplex LP method and a real coded
genetic algorithm (RGA) with and without seeding the random number genera-
tions. As far as the proft is concerned for the farmers, the cost of the feed plays
an important role.
In 1951, Waugh applied linear programming technique developed by
Koopmans (Waugh, 1951) to provide the solution of least cost dairy feed.
Therefore, for the past seven decades conventional linear programming
method was very popular to solve animal diet problems (Rehman, 1984;
Oladokun and Johnson, 2012). In 2001, Tozer applied a multiple-objective
technique to minimize the nutrient excretion from dairy cows through incor-
poration of a nutrient excretion function into ration formulation. In 2002,
Zhang also applied multiple-objective programming to the feed formulation
of a broiler grower ration with the objective of minimizing the nutrient vari-
ance and minimizing the ration cost. Many researchers have introduced the
use of computer software and Excel Solver for solving linear programming
problems (Folayan et al., 2008; Zgajnar et al., 2009; Sebastian et al., 2008;
Radhika and Rao, 2010). Solving diet problems using the LP model math-
ematically only deals with primary technical issues, which can have a weak
relationship with the commercial complication of reducing or increasing
the change between costs and feeding over time. To overcome this problem
various types of methodologies such as goal programming, multi-objective
programming, nonlinear programming etc. have been used for many years
(Rehman and Romero, 1984, 1987; Hertzler, 1987; Afrouziyeh et al., 2010;
Saxena, 2011a). It has been shown that the linear programming approach indi-
cates how best to combine available local ingredients effectively to formulate
the least cost ration for broilers (Al-Deseit, 2009). Ghosh in 2014 combined
LP with goal programming (GP) for feed formulation, displaying the dip in
cost as a major advantage of using the goal-programming approach (Ghosh
et al., 2014).
https://t.me/livestockscience
Feed Formulation Using RGA 47
three different months of pregnancy for different body weights. As the growth
of the fetus increases signifcantly after six months of pregnancy, the nutrient
requirements of cattle at the seventh, eighth and ninth month of pregnancy are
presented in Tables 5.1 and 5.2. The estimation of nutrient requirements dur-
ing late pregnancy requires accurate values for rates of nutrient accretion in
conceptus tissues (ICAR-NIANP, 2013). The ration was formulated using the
following steps:
TABLE 5.1
Nutrient Requirements of Cattle-1, 2 and 3
Body Milk yield Pregnancy
Level weight (kg) (in litres) Fat % (in months)
Cattle 1 500 10 4 7
Cattle 2 500 10 4 8
Cattle 3 500 10 4 9
TABLE 5.2
Selected Feedstuffs
Roughages Concentrates Minerals
https://t.me/livestockscience
48 Livestock Ration Formulation
TABLE 5.3
Nutrient Content of Feedstuffs on DM basis
Feeds DM (%) CP (kg) TDN (kg) Ca (kg) P (kg) Cost (Rs)
Roughage
Paddy straw 90 0.0513 0.4 0.0018 0.0008 5
Co-4 grass 25 0.08 0.52 0.00144 0.0009 3
Maize fodder 25 0.08 0.6 0.0053 0.0014 3
Co-FS 29 sorghum 90 0.07 0.5 0.0012 0.0009 3
Ragi straw 90 0.06 0.42 0.0015 0.0009 3
Concentrates
Maize 90 0.081 0.792 0.00018 0.0027 17
Soya DOC 90 0.42 0.7 0.00018 0.00225 38
Copra DOC 90 0.22 0.7 0.00036 0.009 23
Cotton DOC 90 0.32 0.7 0.00036 0.009 23
Wheat bran 90 0.12 0.7 0.01067 0.00093 17
Gram chunnies 90 0.17 0.6 0.000108 0.00234 12
Cotton seed 90 0.16 1.1 0.003 0.0062 21
Conc. mix Type-I 90 0.22 0.7 0.005 0.0045 17
Calcite 97 0 0 0.36 0 4
MM 90 0 0 0.32 0.06 50
DCP 90 0 0 0.24 0.16 28
Salt 90 0 0 0 0 5
Step-3 (fxing the constraints): To obtain the least cost feed we have some con-
straints for each nutrient, and it is unique for each animal. The minimum and
maximum level of crude protein (CP 9.82–11%), total digestible nutrient (TDN
46.73–51%), calcium (0.38–0.8%), phosphorus (0.23–0.5%), roughage (40–
80%) and concentrates (20–70%) was calculated on total dry matter intake for
each type of cattle. The calculated constraints for each animal are as follows:
°
17
1. i=1
DMI i = 16.75 kg
2. 1.644 kg ˛ °13
i=1 CPi ˛1.8425 kg
3. 7.83kg ˛ °13
i=1 TDN i ˛ 8.5425 kg
4. 0.065 kg ˛ °17
i=1 Cai ˛ 0.134 kg
5. 0.04 kg ˛ °17
i=1 Pi ˛ 0.08375 kg
5
6. 6.7 kg ˛ ° i=1 Roughagesi ˛13.4 kg
7. 3.35 kg ˛ °17
i=6 Concentrates i ˛11.725 kg
2. 1.691 kg ˛ °13
i=1 CPi ˛1.8579 kg
https://t.me/livestockscience
Feed Formulation Using RGA 49
3. 7.793 kg ˛ °13
i=1 TDN i ˛ 8.6139 kg
4. 0.065 kg ˛ °17
i=1 Ca ˛ 0.13512 kg
5. 0.04 kg ˛ °17
i=1 Pi ˛ 0.08445 kg
5
6. 6.756 kg ˛ ° i=1 Roughagesi ˛13.512 kg
7. 3.378 kg ˛ °17
i=6 Concentrates i ˛11.823 kg
1. °17
i=1 DMI i = 17.03 kg
2. 1.738 kg ˛ °13
i=1 CPi ˛1.8733 kg
3. 8.03kg ˛ °13
i=1 TDN i ˛ 8.6853 kg
4. 0.065 kg ˛ °17
i=1 Cai ˛ 0.13624 kg
5. 0.04 kg ˛ °17
i=1 Pi ˛ 0.08515 kg
5
6. 6.812 kg ˛ ° i=1 Roughagesi ˛13.624 kg
7. 3.406 kg ˛ °17
i=6 Concentrates i ˛11.921 kg
Step-4 (fnding results): In this step, we fnd the least cost of “feedstuffs”
after fulflling constraints assigned by solving the linear programming
model with various mathematical tools and then translating the obtained
results as a recommendation to the farmers. The schematic diagram of the
methodology followed is shown in Figure 5.1.
Data Entry
Set Requirement
Maintenance Production
on DM Basis
Converted to
Result out put on DM Basis
Fresh Basis
FIGURE 5.1 Schematic Diagram Representing Methodology Followed for Least Cost Feed
Formulation of Dairy Cattle; Kuntal, AJAVA, 2016.
https://t.me/livestockscience
50 Livestock Ration Formulation
Subjected to:
1. x1 + x2 + x3 + x4 + x5 + x6 + x7 + x8 + x9 + x10 + x11 + x12 + x13 + x14
+ x15 + x16 + x17
1 = 16.75
https://t.me/livestockscience
Feed Formulation Using RGA 51
Subjected to:
1. x1 + x2 + x3 + x4 + x5 + x6 + x7 + x8 + x9 + x10 + x11 + x12 + x13 + x14 + x15
+ x16 + x17 = 16.89
2. 1.691˛ 0.0513x1 + 0.08 x2 + 0.08x3 + 0.07 x4 + 0.06x5 + 0.081x6 + 0.42 x7
+ 0.22x8 + 0.32x9 + 0.12x10 + 0.17 x11 + 0.16x12 + 0.22x13 ˛1.8579
3. 7.93 ˛ 0.4x1 + 0.52 x2 + 0.6 x3 + 0.5x4 + 0.42 x5 + 0.792x6 + 0.7 x7 + 0.7 x8
+ 0.7x
. 9 + 0.7 x10 + 0.6 x11 +1.1x12 + 0.7 x13 ˛ 8.6139
4. 0.065 ˛ 0.0018x1 + 0.00144 x2 + 0.0053x3 + 0.0012 x4 + 0.0015x5
+ 0.00018x
8 6 + 0.00018 x7 + 0.00036 x8 + 0.00036 x9 + 0.01067 x10
+ 0.000108x11 + 0.003 x12 + 0.005x13 + 0.36 x14 + 0.32x15 + 0.24x16
˛ 0.13512
5. 0.04 ˛ 0.0008x1 + 0.0009x2 + 0.0014 x3 + 0.0009x4 + 0.0009 x5 + 0.0027x6
+ 0.00225x7 + 0.009x8 + 0.009x9 + 0.00093x10 + 0.00234x11 + 0.0062 x122
+ 0.0045x13 + 0.06x15 + 0.16 x16 ˛ 0.08445
6. 6.756 ˛ x1 + x2 + x3 + x4 + x5 ˛ 13.512
7. 3.378 ˛ x6 + x7 + x8 + x9 + x10 + x11 + x12 + x13 + x14 + x15 + x16
+ x17 ˛11.823
Subjected to:
1. x1 + x2 + x3 + x4 + x5 + x6 + x7 + x8 + x9 + x10 + x11 + x12 + x13 + x14 + x15
+ x16 + x17 = 17.03
2. 1.738 ˛ 0.0513 x1 + 0.08 x2 + 0.08x3 + 0.07 x4 + 0.06x5 + 0.081x6 + 0.42x7
+ 0.22 x8 + 0.32 x9 + 0.12 x10 + 0.17 x11 + 0.16 x12 + 0.22 x13 ˛1.8733
https://t.me/livestockscience
52 Livestock Ration Formulation
5.2 Implementation
As suggested by Ghosh et al., (2014), we have explored the possibility of using
various techniques including genetic algorithm (GA) to solve the feed formulation
problem (refer to Chapter 4). The success of GA is in its evolution process. More
recently, real coded genetic algorithm (RGA) has been used in numerical optimiza-
tion (Herrera et al., 1995; Ono et al., 2003), and most of the optimization is done in
real space (continuous space). However, constraints are rigid; the use of RGA over-
comes the low resolution of solution weakness of binary coded GA (Baron et al.,
2002; Achiche et al., 2002). The outline of RGA is as follows:
Step-1: Genetic algorithm starts with frst creating the initial population
randomly. In MATLAB 2013, if we set the population size, then the ini-
tial population generated by the algorithm is same as population size by
default, so to solve our linear programming problem we have generated a
500-population size.
Step-2: Sequence of new population and next generation: This creates
a sequence of new population using RGA operators, namely crossover,
mutation and selection. At every step of RGA, the current population is
used to form new offspring that create the next generation. The current
https://t.me/livestockscience
Feed Formulation Using RGA 53
https://t.me/livestockscience
54 Livestock Ration Formulation
TABLE 5.4
Least Cost of Ration by Various Techniques for Cattle-1 on DM Basis
Simplex & EA with RGA with RGA with
Variables GRG nonlinear seeding RNG (1) RNG (17)
TABLE 5.5
Least Cost of Ration by Various Techniques for Cattle-1 on As Fresh Basis (LP Model)
Simplex & RGA without RGA with Real coded GA
Variables GRG nonlinear seeding RNG RNG (1) with RNG (17)
TABLE 5.6
Least Cost of Ration by Various Techniques for Cattle-2 on DM Basis (LP Model)
Simplex & EA with RGA With RGA With
Variables GRG nonlinear seeding RNG (1) RNG (17)
https://t.me/livestockscience
56 Livestock Ration Formulation
TABLE 5.7
Least Cost of Ration by Various Techniques for Cattle-2 on As Fresh Basis (LP Model)
Simplex & RGA without RGA with RGA with
Variables GRG nonlinear seeding RNG RNG (1) RNG (17)
TABLE 5.8
Least Cost of Ration by Various Techniques for Cattle-3 on DM Basis (LP Model)
Simplex & EA with RGA with RGA with
Variables GRG nonlinear seeding RNG (1) RNG (17)
x1 3.85658052 3.5198747 3.1263 3.5056
x2 2.33299069 1.9206234 2.323 2.4336
x3 0.8514 1.1501126 0.8515 0.8515
x4 0.8515 0.8940567 1.82 1.7554
x5 4.2575 1.9046290 4.2393 3.8079
x6 0.8514 1.1700833 0.8515 0.8515
x7 0 0.1522068 0.5362 0.5378
x8 0 0.6827690 0 0
x9 1.91220552 0.3873762 1.1195 1.1239
x10 0.8514 1.4598545 0.8515 0.8515
x11 0 0.8650671 0 0
x12 0 0.1507223 0 0
https://t.me/livestockscience
Feed Formulation Using RGA 57
TABLE 5.9
Least Cost of Ration by Various Techniques for Cattle-3 on As Fresh Basis
Simplex & RGA without RGA with RGA with
Variables GRG nonlinear seeding RNG RNG (1) RNG (17)
https://t.me/livestockscience
58 Livestock Ration Formulation
TABLE 5.10
Net Proft Per Cattle in INR to Farmers for 10 Litres of Milk on As Fresh Basis Each
Day
Milk rate Net proft by Net proft by Net proft by Net proft by
in dairy simplex LP & RGA without RGA with RGA with
Level (INR) GRG nonlinear seeding RNG (1) RNG (17)
and 3 for each type of cattle have 17 variables, which are too complex for fnding
an optimal solution graphically; hence, we used the LP simplex method, which
provides an iterative algorithm to locate the corner points systematically until
we get an optimal solution. A GRG nonlinear algorithm with forward differenc-
ing deals with problems involving decision variables and nonlinear constraints.
Forward differencing uses a single point that is slightly different from current
value to fnd the derivative; hence, while solving LP models in Excel Solver, it
will not compute derivatives again and again, and it continues to estimate the
solution along the straight line instead of recalculating the changing gradients.
When GRG fnds a solution, this means that solver has found a valley for mini-
mizing the objective function after satisfying the Karush-Kuhn-Tucker (KKT)
conditions for local optimality, and there is no other possible solution for deci-
sion variables (feedstuffs) near to current values. The evolutionary algorithm
(EA) technique is also used to solve LP model by seeding the random number
generator. In the present study, the EA method with seeding technique could not
fnd a lower least cost than the other three methods because costlier nutrients
like the crude protein level in fnal feed was higher (1.82 kg) in EA than other
methods (1.644 kg). Similarly, the TDN content is higher in the EA method than
other methods. These could be the reason for higher costs in EA (INR 163.14)
than the other methods (GRG-126.71; LP- 126.71: RGA seeding 1–129.44 and
RGA seeding 17–128.8). It is evident that the EA method had a not so accurate
least cost feed formulation method because it uses only mutation as a parameter
to improve the diversity of the population in every generation. The EA method is
heuristic in nature and uses only mutation as a parameter to improve the diver-
sity of the problem; hence, while solving the problem the constraints are not
satisfed properly and almost all the constraints are violated. If we do not seed
the random number, then we get the wide ranges of solutions in every run most
of the time, but unfortunately in this study we have not achieved the optimal
solution using EA in MS Excel Solver due to the rigidity of the constraints. The
RGA with seeding technique is applied to overcome the limitation of EA. RGA
is also a heuristic technique which works on the principle of survival of best ft
https://t.me/livestockscience
Feed Formulation Using RGA 59
5.2.2 Conclusion
This study addresses the use of RGA as a tool to provide good quality feed mix
to dairy cattle for better health and milk production. All the techniques – viz.
LP simplex, GRG nonlinear, evolutionary algorithm (EA) and real coded genetic
algorithm (RGA) with seeding the random numbers – for least cost ration are per-
forming equally. Hence, it is concluded that RGA (www.scialert.net) will give low-
cost diet formulation for dairy cattle. However, the fxing of constrains and use of
code for making software is also be considered while choosing the techniques for
making least cost feed formulation. Further detailed research with various spe-
cies of animals and with different physiological needs may require to fne-tune
https://t.me/livestockscience
60 Livestock Ration Formulation
the techniques for farmer use. We also tried to cut down cost of TMR without any
shortage of nutrients in the diet.
https://t.me/livestockscience
Feed Formulation Using RGA 61
Animal Husbandry and Fisheries Statistics 2017, the per capita availability of
average milk in Karnataka was 291 grams per day during 2016–17, which is less
than the 12 top milk-producing states in India such as Uttar Pradesh. Karnataka
has only a 4% share in milk production for the year 2016–17. From 2012 to 2016,
the cattle population increased from 1142.62 to 1370.69 (in thousands), with an
estimated milk production of 5718.22 to 6562.15 (in thousands of kg) in which the
average yield per in-milk animal of nondescript/indigenous cows from 2012–13 to
2016–17 in Karnataka was 2.32–2.43 kg/day. Area under fodder crops increased
from 35000 hectares to 36000 hectares in 2006–07, and permanent pastures and
other grazing lands decreased from 930000 hectares to 906000 hectares from
2006–07 to 2013–14 (Basic Animal Husbandry & Fisheries Statistics, 2017).
According to past surveys, it was clear that farmers are not feeding dairy cat-
tle properly due to high feed cost and unavailability of proper feedstuffs (Garg,
2012). Therefore, it is necessary to supply a least cost balanced diet to dairy
cattle, especially during pregnancy and the milking period. Since 1991 many
researchers have studied feeding practices in which small farmers have limited
resources for feeding practices (Leng, 1991: 82). The livestock industry plays an
important role in the development of the Indian economy, as the livestock sector
in agriculture GDP increased from 13.88% to 29.20% from 1990 to 2013. The
livestock sector also amounts to 4% of the national GDP (Department of Animal
Husbandry, Annual Report, 2016–17; Angadi et al., 2016). Hence, when consider-
ing the economic importance and diffculties of Indian farmers, an improvement
in feeding practice is required which results in a least cost feed plan for dairy
cows at different hypothetical conditions. Linear programming (LP) is one of the
commonly utilized strategies pursued by many economical and non-economical
diet formulation programs, but Rehman and Romero described the limitation of
LP while formulating a low-cost diet plan. The inference in the linear program-
ming model restricts the constraints (to be fxed in RHS) and objective func-
tions to a single objective, which implies an application of the goal programming
model which consists of constraints and sets of goals, which are sometimes pri-
oritized (Gupta et al., 2013b: 414–422). The main reason to apply a GP model is
to satisfy all the constraints in terms of goals. Basically, goals can be satisfed
completely or partially, or sometimes goals cannot be met. This problem where
goals cannot be met can be handled by adding +ve and −ve deviational variables
which are defned for each goal individually. After all, the objective function of
a weighted goal programming model optimizes the sum of the total deviation
from set goals where the outcome may lead to a compromise solution between
contradictory goals (Gupta et al., 2013b: 414–422). Zoran Babic et al. applied
a goal programming method to determine an optimal blend of ingredients for
livestock feed in which a GP-based model turns out to be a useful method in
deciding the optimal livestock diet formulation (Babic, 2011). Evolutionary algo-
rithms (EAs) consist of genetic algorithms (GAs), genetic programming and their
hybrid functions (Rehman [Ph.D. thesis] 2014), and an EA highly depends upon
its operators (Koda, 2012: 1–16). Furuya et al. in 1997 used genetic algorithms
in which the ratio of ingredients has evolved. Sahman et al. used a GA to fnd a
least cost diet for livestock, which results in a good solution with few constraints
https://t.me/livestockscience
62 Livestock Ration Formulation
(Şahman et al., 2009: 965–974). Shilpa Jain et al. did the comparative analysis of
a real and binary coded genetic algorithm on a fuzzy time series prediction. The
authors concluded that the real coded GA runs much faster than a binary coded
GA (Shilpa et al., 2013: 299–304). In a previous case study, an LP model of dairy
cows weighing 500 kg which are pregnant at three different months (seventh,
eighth and ninth months) was formulated and solved using LP simplex, GRG
nonlinear, EA and different parameters of a real coded genetic algorithm based
on primary data. This study resulted in “no signifcance difference between tech-
niques” (p > 0.05) and concluded that RGA can be used to formulate the least
cost diet. In this case study we have extended the work to formulate the goal pro-
gramming model of dairy cows, which are pregnant at third trimester, that is, in
the seventh, eighth and ninth month, which required a balanced diet to maintain
health and to produce milk with 4% fat (Kuntal et al., 2016: 594–607). This GP
model is solved using a real coded hybrid genetic algorithm.
1. The cost should be Rs 126.71/- for Cattle-1, Rs 131.82/- for Cattle-2 and
Rs 136.65/- for Cattle-3.
2. Total dry matter intake (DMI) will be 16.75 kg for Cattle-1, 16.89 kg for
Cattle-2 and 17.03 kg for Cattle-3.
3. Crude protein (CP) will be 1.644 kg for Cattle-1, 1.691kg for Cattle-2 and
1.738 kg for Cattle-3.
4. Total digestible nutrients (TDN) will be 8.5425 kg for Cattle-1, 8.6139 kg
for Cattle-2 and 8.6853 for Cattle-3.
5. Calcium (Ca) will be 0.1176 kg for Cattle-1, 0.1223 kg for Cattle-2 and
0.1207 kg for Cattle-3.
6. Phosphorus will be 0.04193 kg for Cattle-1 and 0.04 for Cattle-2 and 3.
7. Roughage will be 12.2858 kg for Cattle-1, 12.2076 kg for Cattle-2 and
12.1495 kg for Cattle-3.
https://t.me/livestockscience
Feed Formulation Using RGA 63
5.3.1.1 GP Model-1
Min ( Z ) =
+p6 (dRough− ) + p7 (dConc− )
2 2
Subjected to:
1. Goal 1(Minimize Least Cost ) : °17 − +
i=1 Ci xi + dcost − dcost = 126.71
= 8.5425 Kg
4. Goal 4 (Maximize Calcium ) : °17 − +
i=1 Cai + dCa − dCa = 0.1176 Kg
8. °17
i=1 xi = 16.75 Kg
5.3.1.2 GP Model-2
Min ( Z ) =
+p6 (dRough− ) + p7 (dConc− )
2 2
Subjected to:
1. Goal 1(Minimize Least Cost ) : °17 − +
i=1 Ci xi + dcost − dcost = 131.8234
https://t.me/livestockscience
64 Livestock Ration Formulation
= 8.6139 Kg
8. °17i=1 xi = 16.89 Kg
5.3.1.3 GP Model-3
Min ( Z ) =
+p6 (dRough− ) + p7 (dConc− )
2 2
Subjected to:
1. Goal 1(Minimize Least Cost ) : °17 − +
i=1 Ci xi + dcost − dcost = 136.65
= 8.6853 Kg
8. °17
i=1 xi = 17.03 Kg
where pi(i = 1,2. . .7) are positive numbers between (0,1) such that p1 > p2 > . . . p7.
https://t.me/livestockscience
Feed Formulation Using RGA 65
https://t.me/livestockscience
66 Livestock Ration Formulation
5.3.3 Results
TABLE 5.11
Least Cost, Deviation, and Constraints Value Solved by Hybrid RGA for Goal
Programming Models
Variables Feedstuffs GP Model-1 GP Model-2 GP Model-3
x1 Paddy straw 0 0 0
x2 Co-4 grass 0 0 0
x3 Maize fodder 12.2802 12.2014 12.1431
x4 Co-FS 29 sorghum 0 0 0
fodder
x5 Ragi straw 0 0 0
x6 Maize 0 0 0
x7 Soya DOC 0 0 0
x8 Copra DOC 0.0480 0 0
x9 Cotton DOC 0.6099 0.7691 0.908
x10 Wheat bran 3.7989 3.9052 3.9654
x11 Gram chunnies 0 0 0
x12 Cotton seed 0 0 0
x13 Conc. mix Type I 0 0 0
x14 Calcite 0 0 0
x15 MM 0 0.0095 0.006
x16 DCP 0.0131 0.0048 0.0074
x17 Salt 0 0 0
https://t.me/livestockscience
Feed Formulation Using RGA 67
5.3.4 Discussion
Table 5.11 shows the results obtained for all the goal-programming models. On
assigning the weights P1 (goal 1: cost), P2 (goal 2: CP), P3 (goal 3: TDN), P4 (goal
4: Ca), P5 (goal 5: P), P6 (goal 6: roughage), P7 (goal 7: concentrate) as 0.9, 0.8,
0.7, 0.6, 0.5, 0.4, 0.3 and solving the GP Model-1 using RGA with hybrid func-
tion, we obtain dcost − = 9.7909, dTDN + = 1.9453, dCa− = 0.0086, dPh− = 0.0132,
dRough− = 0.0056 , dConc− = 0.0074 and the rest of the variables dcost + , dCP + , dCP− ,
dTDN − , dCa+ , dPh+ , dRough + , dConc + as zero. We observe that goals 1, 4, 5, 6 and 7
are overachieved and goal 3 is underachieved, while goal 2 is fully achieved with-
out any deviation, obtaining minimum (Z) = 0.0127. Similarly, on assigning the
same weights P1 (goal 1: cost), P2 (goal 2: CP), P3 (goal 3: TDN), P4 (goal 4: Ca), P5
(goal 5: P), P6 (goal 6: roughage), P7 (goal 7: concentrate) as 0.9, 0.8, 0.7, 0.6, 0.5,
0.4 and 0.3 and solving the GP Model-2 using RGA with hybrid function, we obtain
dcost − = 10.5312 , dCP− = 0.0002 , dTDN + = 1.9789 , dCa− = 0.0115 , dPh− = 0.011 ,
dRough− = 0.0062 , dConc− = 0.0081 and the rest of the variables dcost + , dCP + , dTDN − ,
dCa+ , dPh+ , dRough + , dConc + as zero. Here we also observe that goals 1, 4, 5, 6 and
7 are overachieved and goal 3 is underachieved, while goal 2 is slightly over-
achieved, with dCP− = 0.0002 obtaining minimum (Z) = 0.0132. But on assigning
the same weights P1 (goal 1: cost), P2 (goal 2: CP), P3 (goal 3: TDN), P4 (goal 4: Ca),
P5 (goal 5: P), P6 (goal 6: roughage), P7 (goal 7: concentrate) as 0.9, 0.8, 0.7, 0.6,
0.5, 0.4 and 0.3 and solving the GP Model-3 using RGA with hybrid function, we
obtain dcost − = 11.4146 , dCP− = 0.0001, dTDN + = 2.012 , dCa− = 0.01,dPh− = 0.0096 ,
dRough− = 0.0054 , dConc− = 0.0071 and the rest of the variables dcost + , dCP + , dTDN − ,
dCa+ , dPh+ , dRough + , dConc + as zero. Here it is seen that goals 1, 5, 6 and 7 are over-
achieved and goal 3 is underachieved, while goals 2 and 4 are slightly overachieved,
https://t.me/livestockscience
68 Livestock Ration Formulation
with deviation dCP− = 0.0001 and dCa− = 0.01 obtaining minimum (Z) = 0.0115.
The obtained result may not please the nutritionist completely; hence, he has to
work on overachieved targets. The frst, third, fourth, ffth, sixth and seventh goals
are analysed, and the reason for the overachievement can be searched for in the diet
plan. The choice of the fnal solution relies on the nutritionist, with, in this case,
three different GP-models which represent the diet plan options that a nutritionist
can choose (Babic, 2011). All possibilities are not taken into consideration, as the
LP models developed in (Kuntal et al., 2016) allow introduction of additional con-
straints anytime, which results in a new set of solutions. Some constraints (if added)
can also lead to the result of “no solution”, which means that extra additional con-
straints are so complex that it is necessary to mediate the model by increasing some
of the requirements. However, for better output, we need a further discussion with
a qualifed cattle nutritionist
5.3.5 Conclusion
The study focused on improving the results of the LP model developed by (Kuntal
et al., 2016), formulating it into goal programming models. The GP models are
solved by a real coded genetic algorithm with hybrid function to improve the qual-
ity of feed mix to the dairy cows, where this method confrms it is a useful approach
in fnding the low-cost diet plan for dairy cows at different body conditions. As the
results obtained reveal, that RGA with hybrid function can be applied to formulate
least cost rations; however, fxing the constraints and use of code for making soft-
ware is considered while choosing the technique for making the least cost diet plan.
https://t.me/livestockscience
6
Study of Real Coded Hybrid Genetic
Algorithm to Find Least Cost Ration
for Non-Pregnant Dairy Buffaloes
CONTENTS
6.1 Introduction .................................................................................................. 69
6.2 Data Collection and Data Handling............................................................. 72
6.3 Problem Description .................................................................................... 73
6.3.1 Linear Programming Model............................................................ 74
6.3.2 Goal Programming Model-1 ........................................................... 75
6.3.3 Goal Programming Model-2 ........................................................... 76
6.3.4 Goal Programming Model-3 (Auxiliary Approach) ....................... 76
6.4 Development of Real Coded Hybrid Genetic Algorithm ............................ 77
6.4.1 Objective Function, Decision Variables, Representation
of Plots ............................................................................................. 77
6.4.2 Crossover, Selection, Mutation and Elitism Operators ................... 78
6.4.3 Global vs. Local Minima................................................................. 78
6.5 Results on Dry Matter and As-Fresh Basis.................................................. 79
6.6 Discussion .................................................................................................... 80
6.7 Conclusion .................................................................................................... 86
6.1 Introduction
In the Mandya district of Karnataka, the cost of milk per litre was more in the case
of buffaloes compared to local cows due to the high fat content and high nutritive
value of buffalo milk compared to its counterpart. Based on earlier research, it was
clear that the productivity of the buffaloes maintained by different dairy farms
was lower. Therefore, there is a need to focus on two important aspects of dairy
farming: one to increase the milk productivity of buffaloes and the other one to
reduce the diet cost by upgrading scientifc dairy farming practices. Even though
there are many predefned techniques used for animal diet formulation, a strong
application of soft computing is always preferred to enhance the quality of the
solution in which the rigidity in the functions as well as constraints can be easily
handled by a linear programming program (LPP). Therefore, to meet the nutrient
requirements at the lowest cost, we have developed a real coded hybrid genetic
algorithm (RHGA) for formulating the least cost ration for dairy buffaloes, tak-
ing the hybridization of RGA and the “fmincon” of Excel Solver. This technique
is better than old conventional techniques in the sense that it does not break if
the inputs are modifed and provides better results over complex problems even
DOI: 10.1201/9781003231714-6 69
https://t.me/livestockscience
70 Livestock Ration Formulation
https://t.me/livestockscience
Study of RHGA to Find Least Cost Ration 71
and should be of low cost. Since 1991, many researchers have studied the feeding
practices of dairy cattle, in which small farmers have limited resources for feed-
ing and do not have proper knowledge or resources to provide a low-cost balanced
ration (Leng, 1991). Study also reveals that in most of the country, animal diet is
imbalanced by over-feeding energy, protein and minerals. Hence, there is a large
scope for improvement in the nutritional status by adopting improved feeding prac-
tices (Mudgal et al., 2003). In addition, the dairy sector in India is highly dependent
upon buffalo milk because of its contribution to total milk production (49%), as it
is known that buffalo milk is rich in fat and solid content. The livestock industry
plays a major role in the development of the Indian economy as the share of the
livestock sector in agricultural GDP increased from 13.88% in 1980–81 to 29.20%
in 2012–13 (Reddy et al., 2016) According to the Annual Report of the Department
of Animal Husbandry, Dairying & Fisheries Ministry of Agriculture, Government
of India (2015–16), livestock is one of the most important contributors in the Indian
economy. As per the annual Report of Dept. of Animal Husbandry, Dairying and
Fisheries (2016–17) India stands frst in the world in milk production (155.5 million
tonnes). Livestock also contribute to 4% of the national GDP (Angadi et al., 2016).
A feld study on the effects of feeding a balanced ration on milk production con-
cerning nitrogen use effciency (NUE) was conducted on 7090 lactating cows and
4534 lactating buffaloes, in which using local feed resources, a balanced least cost
ration can be obtained (Garg et al., 2016). Therefore, by considering the economic
importance and diffculties of Indian farmers and the importance of buffalo rear-
ing, an improvement in feeding practices is required which results in effective milk
production by optimizing the utilization of available feedstuffs and also attempts to
minimize the feed cost of rations. Linear programming (LP) is a standout amongst
the most normally utilized strategies followed in numerous business and non-busi-
ness bolster plan programs. However, Rehman and Romero (1984) call attention to
the fact that LP has numerous restrictions while detailing apportions, practically
speaking. The presumptions in the LP strategy require objectives to be single and
requirements to be settled on the right-hand side (RHS) (Gupta et al., 2013b). This
implies the diminishment of a goal programming (GP) model comprised of limi-
tations and a set of goals which are organized a few times. The goal of GPs is to
discover the arrangement which fulfls the limitations and approach the expressed
objectives of separate issues. Hypothetically, goals could be fulflled totally,
incompletely or, in some exceptional cases, not at all. This possibility is estimated
utilizing positive and negative deviation factors that are characterized for every
objective independently, normally known as fnished or under-accomplishment
of the objective. Since the target capacity of the GP plan limits the whole of the
aggregate deviation from set objectives, the acquired outcome may yield a trade-
off arrangement between opposing objectives (Gupta et al., 2013a). Evolutionary
algorithms consist of genetic algorithm, genetic programming and their hybrid
functions (Rehman [Ph.D. Thesis] 2014), and EA highly depends upon its operators
(Koda, 2012). Furuya et al. in 1997 used genetic algorithms in which the ratio of
ingredients evolved. In this paper, he used a new form of crossover and mutation
in which mutation was generated by a combination of uniform random number and
normal distribution random numbers, and the mutation rate was very high. He also
https://t.me/livestockscience
72 Livestock Ration Formulation
solved the nonlinear constraint which showed that EA is good technique for a diet
problem (Tozer and Stokes, 2001). Sahman et al. (2009) used GA to fnd least cost
diet for a livestock, which results in good solution with few constraints.
In a nutshell we can say that appropriate feeding is the most important feature
in dairy buffalo management, as the feed costs account for more than half of the
total milk production (Clark and Davis, 1980). The quantity of feed that is utilized
for milk production should be optimum as well as satisfy the nutrient requirement
for the animal, to increase the proftability of dairy farmer. An LP model develops
results with a single objective of minimizing the feed cost and rigid constraints,
while converting an LP model to a GP model gives a lot of fexibility in decision
making. The rigid constraints can be considered as a goal with a specifc target with
permissible deviations, and the solution can be seen in terms of underachievement,
overachievement or fully achieved. Fixing the priorities of each goal also leads to
multiple solutions, and a judicial decision can be made depending upon the prefer-
ence of the dairy farmers. Since many researchers in the past addressed the issue
of diet problems using EA, in this chapter, we made an effort to formulate a low-
cost ration for non-pregnant buffaloes of body weight 450 kg at the third lactating
period, with 10 litre milk production containing 6% fat, which was solved using
an RHGA. The study was further extended to formulate nonlinear weighted goal
programming models by considering the objective as the square root of the sum of
the squares of the deviations.
TABLE 6.1
Nutrient Requirements for Non-Pregnant Buffalo Weighing 450 kg and Yielding
10 Litres of Milk with 6% Fat
Total DMI (kg) CP (kg) TDN (kg) Ca (kg) Phosphorus (kg)
min–max min–max min–max min–max min–max
16.42–17.24 1.7158–1.8016 7.9857–9.1835 0.0680–0.0748 0.0270–0.0405
https://t.me/livestockscience
Study of RHGA to Find Least Cost Ration 73
TABLE 6.2
Nutrient Content of Feedstuffs on DM Basis
DM CP TDN Ca P Cost Ci
Feedstuffs (in %) (in kg) (in kg) (in kg) (in kg) (in Rs)
https://t.me/livestockscience
74 Livestock Ration Formulation
FIGURE 6.1 Methodology to Calculate Least Cost Balanced Ration for Non-Pregnant Dairy
Buffalo; Kuntal, Adv in Intelligent System & Computing, 2017.
by NIANP, Bangalore, as per the ICAR 2013 and NRC 2001 standard. A linear
programming model is introduced for the specifed condition of buffalo. Although
there is a well-defned method to solve LPP, an evolutionary algorithm (EA) is
preferable, as these algorithms effciently deal with the mathematical rigidity of the
constraints, can handle large dimension problems and also give multiple solutions.
Since the number of variables represented as feedstuffs are greater in number and
the constraints are very rigid, we opted to develop an RHGA to solve the ration
formulation problem for buffaloes. Various versions of this algorithm were created,
and a comparative study was planned to obtain the appropriate method which can
be used by the farmers at farm level.
Subjected to:
21
1. 16.42Kg ° ˜ i=1 x i °17.42Kg
21
2. 1.7158Kg ° ˜ i=1 CPi °1.8016Kg
https://t.me/livestockscience
Study of RHGA to Find Least Cost Ration 75
21
3. 7.9857Kg ° ˜ i=1 TDN i ° 9.1835Kg
21
4. 0.0680Kg ° ˜ i=1 Ca i ° 0.0748Kg
21
5. 0.0270Kg ° ˜ i=1 Ph i ° 0.0405Kg
6. 6.568Kg ° ˜8i=1 Rough i °13.136Kg
21
7. 3.224Kg ° ˜ i=9 Conc i °11.494Kg
Search Space:
0.806 ° x1 ° 4.03, 0.806 ° x2 ° 4.03, 0.806 ° x3 ° 4.03, 0.806 ° x4 ° 4.03, 0.806
8
° x5 ° 4.03, 0.806 ° x6 ° 3.224, 0.806 ° x7 ° 3.224, 0.806 ° x8 ° 4.03,
0.806 ° x9 ° 3.224, 0 ° x10 ° 4.03, 0 ° x11 ° 4.03, 0 ° x12 ° 3.2240,
0 ° x13 ° 3.224, 0.1612 ° x14 ° 0.806, 0.1612 ° x15 ° 0.806, 0.0806 ° x16 ° 0.806,
.
0.806 ° x17 ° 3.224, 0 ° x18 ° 0.16120, 0 ° x19 ° 0.0806, 0 ° x20 ° 0.03224,
0.12896 ≤ x21 ≤ 0.1612
Min ( Z ) =
+p6 (dPh + ) + p7 (dRough + ) + p8 (dConc + )
2 2 2
Subjected to:
= 9.1835 Kg
5. Goal 5(minimize Calcium ) : ˜ i=1
21
Cai + dCa− − dCa + = 0.0748 Kg
6. Goal 6 (minimize Phosphorus) : ˜ i=1
21
Phi + dPh− − dPh + = 0.0405 Kg
7. Goal 7(minimize Roughages) : ˜ i=1
21
Roughi + dRough− − dRough +
= 13.13
36 Kg
8. Goal 8(minimize Concentrates) : ˜ i=1
21
Conci + dConc− − dConc + = 3.284
40 Kg
https://t.me/livestockscience
76 Livestock Ration Formulation
TABLE 6.3
Priority Values for GP Model-1
Priorities Goals Value
TABLE 6.4
Priority Values for GP Model-2
Priorities Goals Value
Subjected to:
Min ( Z ) =
+p6 (dRough− ) + p7 (dConc− )
2 2
Subjected to:
https://t.me/livestockscience
Study of RHGA to Find Least Cost Ration 77
8. ˜i21
=1 xi = 16.42 Kg
where pi (I = 1,2. . .7) are positive numbers between (0,1) such that p1 > p2 > . . . p7
https://t.me/livestockscience
78 Livestock Ration Formulation
https://t.me/livestockscience
Study of RHGA to Find Least Cost Ration 79
in the wider range by increasing the diversity of the initial population by manip-
ulating the initial population. However, we have rigid constraints and bounds,
so we want to search the point in the specifed lower and upper bounds only.
“Fmincon” is a gradient-based search technique which works on a problem with
continuous constraints and objective functions and must have a frst derivative. In
MATLAB, “fmincon” uses a sequential quadratic programming (SQP) sub-prob-
lem where SQP methods are a representation of nonlinear programming methods.
Schittkowski (Singh and Singh, 2015) has executed a version that surpasses every
other tested method in terms of effciency and accuracy of the solution for a large
number of test problems (tel.archives-ouvertes.fr). Based on previous work (Biggs,
1975; Han, 1977; Powell, 1978a), it can be seen that the method mimics Newton’s
method for a constrained problem. At every generation, an initial judgment is
made of the Hessian of the Lagrangian function using a quasi-Newton method,
which is used to generate a quadratic programming (QP) problem whose solution
is a benchmark which is used to form a search direction for a line search prob-
lem (tel.archives-ouvertes.fr). A major overview of SQP is discussed in (Fletcher,
1987; Gill et al., 1981; Powell, 1978b; Hock and Schittkowski, 1983). In light of
this, we have used the SQP algorithm for “fmincon” Solver, where the SQP algo-
rithm (identical to SQP–legacy algorithm) is the same as the active-set algorithm
but has a different implementation. An interior point algorithm can also be used,
but SQP has faster execution time and less memory usage then the SQP-legacy and
interior point algorithms. Nocedal and Wright explain the basic SQP algorithm
(Nocedal and Wright, 2006).
(Continued )
https://t.me/livestockscience
80 Livestock Ration Formulation
TABLE 6.5
(Continued)
Minimum value obtained in 10 runs for 5000 generations
RGA with
RGA without RGA without RGA with crossover hybrid
Feedstuffs crossover mutation and mutation function
6.6 Discussion
The results obtained by various techniques – viz. RGA without crossover, RGA with-
out mutation, RGA with crossover and mutation and RGA with hybrid function – for
least cost ration for dairy buffalo (which is run 10 times up to 5000 generations)
are presented in Table 6.5. The LP model for dairy buffalo has 21 variables, which
are too complex for fnding an optimal solution. RGA is also a heuristic technique,
which works on a principle of survival of best ft and tournament selection. Adaptive
feasibility mutation and heuristic crossover are used along with elitism to solve the
LP model for dairy buffalo. Although RGA without crossover, RGA without muta-
tion and RGA with crossover and mutation give a near-optimal answer, we provide
a solution using RGA with hybrid function to avoid the solution being stuck in local
minima. As farmers need a total mixed ration to feed dairy buffalo on as fresh basis,
we have converted the least cost and total mixed ration obtained by all techniques to
as fresh basis, and the results are given in Table 6.6. As per Table 6.5, this category
of animals requires about 16.42 kg of dry matter from various kinds of feeds, which
should contain 1715.8 g of protein, 9183.5 g of TDN, 74.8g of calcium and 40.5 g
of phosphorus. These nutrient requirements can be met from 13.13 kg of roughage
and 3.28 kg of concentrates on dry matter, amounting to Rupees 101.6703 per day
https://t.me/livestockscience
Study of RHGA to Find Least Cost Ration 81
per cattle on dry matter basis. As fresh basis is a feed nutrient content with mois-
ture included. After converting to as fresh basis, feedstuffs required for dairy buf-
falo are approximately 36.3 kg per day, amounting to Rs 4.52 per kg using the RGA
with hybrid function. When the expected nutrient requirement solution was sought
using RGA without crossover, without mutation and with crossover and mutation,
the least cost ration obtained was Rs 4.59 per kg, Rs 4.93 per kg and Rs 4.67 per kg,
respectively. The detailed analysis of the ration is showed in Tables 6.5 and 6.6 which
exactly meets the requirements of dry matter, CP, TDN, calcium and phosphorus. The
requirements of roughage and concentrates are also met within the permissible range.
In addition, results obtained by GP Model-1, 2 and 3 are shown in Tables 6.7, 6.8 and
6.9, which reveals that from GP-1 goals 5 and 8 are fully achieved and goals 1, 2, 3, 4,
6 and 7 are underachieved. The least cost obtained by GP-1 is Rs 100.79 on DM basis,
but DMI and CP were not satisfed. Whereas by GP Model-2, our high priorities,
TABLE 6.6
Least Cost Ration Formulated by Various Techniques for Non-Pregnant Dairy Buffalo
on “As Fresh Basis” (LP Model)
RGA with RGA with
RGA without RGA without crossover and hybrid
Feedstuffs crossover mutation mutation function
https://t.me/livestockscience
82 Livestock Ration Formulation
TABLE 6.7
Least Cost and Deviation Value Solved by Hybrid RGA for GP Model-1
Feedstuffs Values
https://t.me/livestockscience
Study of RHGA to Find Least Cost Ration 83
Feedstuffs Values
TDN 7.6663
Calcium (Ca) 0.0748
Phosphorus (P) 0.0287
Roughage 10.8248
Concentrates 3.2840
Least cost on DM basis 100.7965
As fresh basis
TMR on as fresh basis in kg 26.91883
Least cost in Rs on as fresh basis 135.0133
Least cost (in Rs per kg) 5.015569
TABLE 6.8
Least Cost and Deviation Value Solved by Hybrid RGA for GP Model-2
Feedstuffs Values
(Continued )
https://t.me/livestockscience
84 Livestock Ration Formulation
TABLE 6.8
(Continued)
Feedstuffs Values
dCP− 0.0032
dCP+ 0.0997
dTDN− 0.2440
dTDN+ 1.5679
dCa− 0.0055
dCa+ 0.2868
dPh− 0.0663
dPh+ 0.7203
dRough− 0.0216
dRough+ 0.0182
dConc− 0.2544
dConc+ 0.0095
DMI 16.4200
CP 1.7158
TDN 9.1835
Calcium (Ca) 0.0748
Phosphorus (P) 0.0405
Roughage 13.1359
Concentrates 3.2841
Least cost on DM basis 101.6090
As fresh basis
TMR on as fresh basis in kg 30.62294
Least cost in Rs on as fresh basis 143.947
Least cost (in Rs per kg) 4.700626
TABLE 6.9
Least Cost and Deviation Value Solved by Hybrid RGA for GP Model-3
Feedstuffs Values
https://t.me/livestockscience
Study of RHGA to Find Least Cost Ration 85
Feedstuffs Values
goals 1 and 2, are completely satisfed and the rest of the goals are either under-
achieved or overachieved, but the least cost obtained was Rs 100.6090 per day per
cattle on dry matter basis, which is same as the LP model. The amount of feedstuffs is
much less compared to the LP model, which is truly an advantage to farmers because
of the limitation of feedstuff. From a farmer’s point of view, the TMR required for
https://t.me/livestockscience
86 Livestock Ration Formulation
the farmer by the RGA hybrid function is 26.9 kg per day, amounting to Rs 5.01
per kg, whereas by GP Model-2, the TMR required is 30 kg per day, amounting to
approximately Rs 4.7 per kg, which is less compared to the RGA hybrid function.
However, for a better output for farmers, we need a further discussion with a qualifed
cattle nutritionist. For GP Model-3, we observe that on assigning the weights P1 (goal
1: cost), P2 (goal 2: CP), P3 (goal 3: TDN), P4 (goal 4: Ca), P5 (goal 5: P), P6 (goal 6:
roughage), P7 (goal 7: concentrate) as 0.9, 0.8, 0.7, 0.6, 0.5, 0.4 and 0.3 and solving the
GP Model using RGA with hybrid function, we obtain dcost− = 0.0021, dCP+ = 0.4508,
dTDN+ = 0.0901, dCa+ = 0.9891, dPh+ = 0.0499, and rest of the deviational variables dcost+,
dCP−, dTDN−, dCa−, dPh−, dRough−, dRough+, dConc−, dConc+ are zero. We observe that goal 1
is overachieved, goals 2, 3, 4 and 5 are underachieved and goals 6 and 7 are fully
achieved without any deviation with minimum (Z) = 0. The results obtained in this
work are better than the results obtained for GP Model-1. On comparing the results
obtained in both the models, dcost−, dDM−, dCP−, dTDN−, dPh−, dRough− are overachieved,
while dCa−, dCa+, dConc−, dConc+ are fully achieved, due to which all constraints except
calcium and concentrates are satisfed with the least cost of Rs 100.605/- on DM
basis. The reason for not satisfying the constraints by the goal programming model is
adding the deviation variables to dry matter basis; that is, dry matter intake is one of
the fxed constraints, and hence it has to be treated as a constraint rather than treat-
ing it simply as a goal. In the present work, after considering all the loopholes, a new
goal-programming model has been developed which gives an improved solution in
which almost all the constraints are satisfed including our high priority goals (least
cost and dry matter intake). The obtained result was not completely accepted by the
nutritionist, as the reasons for underachieved and overachieved targets need to be
analysed, because the choice of a fnal solution depends on the nutritionist, and there
is a need of further discussion with the nutritionist for better output.
6.7 Conclusion
The present study addresses the use of a real coded genetic algorithm with hybrid
function (RHGA) as a tool to provide a good quality feed mix to the dairy buf-
falo for better health and milk production. All the techniques – viz. RGA with-
out crossover, RGA without Mutation, RGA with crossover and mutation and RGA
with hybrid function – for the least cost ration are performing equally. RGA without
crossover and RGA without mutation operator provide near-optimal answers, but
the solutions seem to be get stuck in local minima; hence, it is proved that the real
coded genetic algorithm with hybrid function (RHGA) provides an optimal solution,
and this method can be used effectively for a low-cost diet plan for dairy buffalo.
Nutrient requirements of buffalo are calculated by an Excel-based computer pro-
gram developed by NIANP, Bangalore, as per the Indian Council of Agricultural
Research–ICAR 2013 and NRC 2001 standard. Application of the RHGA in solv-
ing GP Model-1, 2 and 3 for the ration formulation problem of Indian buffaloes
in the Mandya district of Karnataka is found to be very effcient. Further detailed
research with various species of animals and with different physiological needs may
be required to fne-tune the techniques for farmers’ use. In addition to that, we are
able to cut down the cost of TMR without any nutrient defciency in the diet.
https://t.me/livestockscience
7
Conclusion
The main purpose of formulating the diet problem is to fnd the combination of
feed ingredients which satisfes the entire daily nutritional requirements of dairy
animals with minimum cost. As the farmer’s objective is to reduce the cost of
ration and to maximize the proft, since spending more on rations will decrease the
proftability of farmer, this can be achieved by scientifc and economic effciency.
This study is the proof that optimization of the ration at a cheaper rate is pos-
sible at different body conditions of dairy cows and buffaloes by combining linear
and goal programming (LP and GP) with priority functions and using a heuristic
technique called a real coded genetic algorithm (RGA). The entire idea was frst
tested with a nonlinear model for Sahiwal cows with the objective of maximizing
the milk yield by Pratiksha Sexena using a binary coded genetic algorithm. The
effectiveness of the binary coded genetic algorithm was also tested on three linear
programming model and a nonlinear goal programming model extracted from the
work of Shrabani Gosh, where these mathematical models are developed for differ-
ent conditions, viz: if a fully grown cow having a body weight of 500 kg:
Later, three linear programming (LP) models and nonlinear programming (NLP)
models are formulated for the least cost diet of dairy cows of body weight 500 kg,
where the cows during pregnancy at third trimester need a balanced ration for body
maintenance, milk production (with a minimum of 4% of fat content) and for fetus
growth. These models are solved by an RGA (with and without seeding the random
number generation), and the results are compared with Excel Solver tools, namely
GRG nonlinear, evolutionary algorithm (EA) and simplex LP. Further, the study
has been extended to the development of linear and goal programming models of
non-pregnant dairy buffalo of body weight 450 kg, which need a ration to produce
10 litres of milk with 6% fat content. It has been observed that in all these case
studies, reductions of ration cost is possible by heuristic techniques as compared to
the old conventional method, irrespective of the rigidity of constraints. For better
diet formulation, the minimum and maximum range (in percentage) of crude pro-
tein (CP), total digestible nutrients (TDN), dry matter (DM), calcium, phosphorus,
roughage and concentrates are introduced in both LP and GP models based on ICAR
2013 and NRC 2001 standards. The RGA and hybrid genetic algorithm (RHGA)
DOI: 10.1201/9781003231714-7 87
https://t.me/livestockscience
88 Conclusion
https://t.me/livestockscience
Conclusion 89
applied effectively for fnding the least cost diet plan. In addition, after analysing
the approach applied by Shrabani Ghosh, it was found that minerals are not used
while formulating the diet plan. But there are three different body conditions of
animals, which defnitely requires additional nutrient supplements called minerals,
and this supplement cannot be obtained from feedstuffs.
In Chapter 5, we took into account this important aspect of including minerals
as essential nutrient requirement for the animal while formulating the diet plan for
cattle in the Mandya district of Karnataka. Using the primary data of the Mandya
district of Karnataka, taken from NIANP Bangalore, three different linear mod-
els were formulated for dairy cattle with body weight 500 kg and a 10 litre milk
yield with 4% of fat content who were in the seventh, eighth and ninth months of
pregnancy. The constraints for these LP models are fxed by giving a minimum
and maximum percentage range on dry matter basis, which was not considered
by researchers earlier. This feature makes the model easy to understand, but at the
same time the model becomes rigid. The minimum and maximum level of CP is
9.82–11%, TDN is 46.73–51%, calcium is 0.38–0.8%, phosphorus is 0.23–0.5%,
roughage is 40–80% and concentrate is 20–70%, as calculated on total DMI for
each type of cattle. These LP models are solved using LP simplex, GRG nonlinear,
evolutionary algorithm (EA) and real coded genetic algorithm (RGA) with seed-
ing the random numbers. The LP simplex method provides an iterative algorithm
to locate the corner points systematically until we get an optimal solution. The
GRG nonlinear algorithm with forward differencing deals with problems involving
decision variables and nonlinear constraints. Forward differencing uses a single
point that is slightly different from the current value to fnd the derivative. Solving
LP models in Excel Solver, it will not compute derivatives again and again, and it
continues to estimate the solution along the straight line instead of recalculating the
changing gradients. When GRG fnds a solution, this means that Solver has found
a valley for minimizing the objective function after satisfying the Karush-Kuhn-
Tucker (KKT) conditions for local optimality, and there are no other possible solu-
tions for decision variables (feedstuffs) near to current values. EA is also used to
solve the LP model by seeding the random number generator. The results obtained
by various techniques for LP models, are presented in Tables 5.4, 5.6 and 5.8 for
Cattle-1, 2 and 3 on DM basis. LP models 1, 2 and 3 have 17 decision variables
and eight constraints with minimum and maximum range. In this study, the EA
method with seeding technique could not fnd least cost because costlier nutrients
like the crude protein level in fnal feed was higher (1.82 kg) in EA as compared to
other method (1.644 kg). Similarly, TDN content is also higher in the EA method
than other methods. These could be the reasons for higher costs (INR 163.14)
obtained using EA compared to other methods (GRG–126.71; LP–126.71: RGA
seeding 1–129.44 and RGA seeding 17–128.8). It is evident that the EA method
is not so accurate for least cost feed formulation because it uses only mutation as
a parameter to improve the diversity of the population in every generation. EA is
heuristic in nature but uses only mutation as one parameter to improve the diversity
of the problem. Hence, while solving the problem the constraints are not satisfed
properly and almost all the constraints are violated. If we do not seed the random
number, then we get the wide ranges of solutions in every run most of the time, but
https://t.me/livestockscience
90 Conclusion
unfortunately in this study we did not achieve the optimal solution using EA in MS
Excel Solver due to the rigidity of the constraints.
An RGA with seeding technique is developed to overcome the limitation of EA.
RGA is also a heuristic technique which works on the principle of the survival
of best ft and tournament selection. Adaptive feasibility mutation and heuristic
crossover is used along with elitism to solve the LP model for Cattle-1, 2 and 3.
Though the seeding technique gives near-optimal answers, we prefer to provide
the wide ranges of solutions to farmers in which the values of feedstuffs change in
every run where all the constraints will satisfy. As a farmer needs a total mixed
ration to feed the cattle on as fresh basis, we have converted the least cost and total
mixed ration obtained by all techniques to as fresh basis, and the results are given
in Tables 5.5, 5.7 and 5.9. As per the ftment in Table 5.4, an animal of each cat-
egory requires about 16.75 kg of dry matter from various feed ingredients, which
should contain 1644 g of protein, 8542 g of TDN, 117.6 g of calcium and 42 g of
phosphorus, and these nutrients can be met from 12.29 kg of roughages and 4.464
kg of concentrate on dry matter. The corresponding TMR cost on dry matter basis
is Rs 126.70, that is, Rs 7.56 per kg. As fresh basis is a feed nutrient content with
moisture included. After converting to as fresh basis, the feedstuff requirement for
Cattle-1 is approximately 30 kg/day, amounting to Rs 5.83/kg using the GRG non-
linear and simplex LP technique. When the expected nutrient requirements were
tested using an RGA without seeding and with seeding (RGA1 and RGA17), the
least cost ration obtained was Rs 6.05/kg, Rs 5.96/kg and Rs 5.99/kg, respectively.
The detailed analysis of rations showed (in Tables 5.4 and 5.5) that this method
exactly met the requirement of dry matter, TDN, CP, calcium and phosphorus, and
roughage–concentrate was also well within the permissible range. Similar analy-
sis has been done for Cattle-2 (see Tables 5.6 and 5.7), where the least cost ration
obtained by various techniques for as fresh basis is Rs 6.08, Rs 6.25, Rs 6.40 and
Rs 7.20 per kg by LP simplex, GRG nonlinear, RGA (1) and RGA (17), respectively.
For Cattle-3 (see Tables 5.8 and 5.9), the TMR cost for as fresh basis turns out to
be Rs 6.38, Rs 6.59, Rs 6.50, Rs 6.50 per kg, respectively. Table 5.10 shows the net
proft that farmers can make per cattle in Indian rupees for 10 litres of milk each
day on as fresh basis by all techniques. A one-way ANOVA test at 5% level of
signifcance has been performed for the “Null hypothesis: there is no signifcant
difference between techniques”. The test reveals that since the P value is greater
than 0.05, there is no signifcance difference between the techniques. Hence, it is
proved that the real coded genetic algorithm (RGA) method can be used for ration
formulation to fnd least cost feedstuffs in dairy cattle.
A further linear model was converted to formulate the goal programming model
with priority functions. By applying the GP model in this study, the cost of rations
could be reduced to a reasonable extent satisfying the exact nutrient requirement.
The cost of diet can be reduced by giving frst priority to underachievement of least
cost and the other priority to overachievement of other nutrients such as CP, TDN,
Ca, P, concentrates and roughages on a dry matter basis. Basically, many researchers
will try to adjust the dry matter level or change the ratio of concentrates–roughage
to reduce the cost of diet by using the GP model, but we have used the same range
for constraints which was used earlier for the LP model. Table 5.11 shows the result
https://t.me/livestockscience
Conclusion 91
obtained for GP Model-1, 2 and 3 by the real coded genetic algorithm after assign-
ing the weights on P1 (goal 1: cost), P2 (goal 2: CP), P3 (goal 3: TDN), P4 (goal 4:
Ca), P5 (goal 5: P), P6 (goal 6: roughage) and P7 (goal 7: concentrate) as 0.9, 0.8, 0.7,
0.6, 0.5, 0.4 and 0.3. The results obtained reveal that for GP Model-1, using RGA
with hybrid function, we obtain dcost− = 9.7909, dTDN+ = 1.9453, dCa− = 0.0086, dPh− =
0.0132, dRough− = 0.0056, dConc− = 0.0074 and the rest of the variables dcost+, dCP+,
dCP−, dTDN−, dCa+, dPh+, dRough+, dConc+ as zero. Goals-1, 4, 5, 6 and7 are overachieved
and goal-3 is underachieved, while goal-2 is fully achieved without any deviation,
obtaining minimum (Z) = 0.0127, which results in a cost of Rs 116.9191/- per kg, per
cattle. For GP Model-2, the deviational values are dCP− = 10.5312, dCP− = 0.0002,
dTDN+ = 1.9789, dCa− = 0.0115, dPh− = 0.011, dRough− = 0.0062, dConc− = 0.0081, and the
rest of the variables dcost+, dCP+, dTDN−, dCa+, dPh+, dRough+, dConc+ as zero, where goals-
1, 4, 5, 6 and7 are overachieved and goal-3 is underachieved, while goal-2 is slightly
overachieved with dCP− = 0.0002 obtaining minimum (Z) = 0.0132, amounting to
Rs 121.2922/- per kg, per cattle on DM basis. For GP Model-3, the deviational
values are dCP− = 11.4146, dCP− = 0.0001, dTDN+ = 2.012, dCa− = 0.01, dPh− = 0.0096,
dRough− = 0.0054, dConc− = 0.0071 and the rest of the variables dcost+, dCP+, dTDN−, dCa+,
dPh+, dRough+, dConc+ as zero where goals-1, 5, 6 and7 are overachieved and goal-3 is
underachieved, while goals-2 and 4 are slightly overachieved with deviation dCP− =
0.0001 and dCa− = 0.01 with minimum (Z) = 0.0115, amounting to Rs 125.2354/- per
kg, per cattle on DM basis. Overall, the cost of Cattle-1, 2 and 3 obtained by the
goal programming approach is cheaper by 7.72% for Cattle-1, 7.98% for Cattle-2
and 8.34% for Cattle-3 in comparison with the results obtained by LP model-1, 2
and 3 on dry matter basis, respectively.
The results obtained by the GP model do not completely satisfy the decision
maker; therefore, the decision maker needs to work on the overachieved targets.
First, the third, fourth, ffth, sixth and seventh goals are analysed, and the reason
for the overachievement can be searched for in the diet plan. All possibilities are not
considered, as the LP model developed earlier allows the introduction of additional
constraints anytime, which results in a new set of solutions, while some constraints
(if added) can also lead to “no solution”, which means that additional constraints
are too complex and it is necessary to mediate in the model by increasing some of
the requirements. However, for better output, we need a further discussion with a
qualifed cattle nutritionist.
Further, in Chapter 6, three goal programming models for non-pregnant buffalo
weighing 450 kg and yielding 10 litres of milk with 6% fat are developed on a dry
matter basis. In GP Model-1, all eight goals are considered for minimization. In GP
Model-2, only two goals, least cost and dry matter, are considered for minimiza-
tion, and in GP Model-3, other than least cost, all other goals are of a maximizing
nature. These GP models are solved by an RGA with the hybrid function “fmin-
con”, and the results obtained reveal that an RHGA can effectively be used to econ-
omize the total mixed ration cost such that the feed requirements of the animals
are met without any nutrient defciency. Initially, the linear model is considered and
solved by real coded GA with different parameters, and due to the large dimension
of the problem and rigidity in constraints, we opted to develop an RHGA to solve
the ration formulation problem for buffaloes. This category of animal requires 16.42 kg
https://t.me/livestockscience
92 Conclusion
of dry matter from various kinds of feedstuffs, which should contain 1715.8 g of
protein, 9183.5 g of TDN, 74.8 g of calcium and 40.5 g of phosphorus. As per the
LP model solved using hybrid GA, this requirement can be met from 13.13 kg of
roughage and 3.28 kg of concentrates on dry matter. Its corresponding cost on dry
matter basis is Rs 101.6703/- per kg, per buffalo. As fresh basis is a feed nutrient
content with moisture included for total mixed ration (TMR); hence, it requires
approximately 36.3 kg TMR, amounting to Rs 4.5/- per kg. By GP Model-1 the
approximately cost of ration is Rs 100.7965/-on DM basis, which corresponds to
26.9 kg TMR, amounting to Rs 5.01/- per Kg, where the goals 1, 2 3, 4, 6 and 7 are
overachieved and goals 5 and 7 are fully achieved. This GP model corresponds to
least cost, but the constraints such as least cost, CP, TDN, P, roughage and concen-
trates are decreased by 14%, 17.4%, 16.5%, 29.13% and 17.594%, and calcium and
phosphorus are achieved completely, 100%. Therefore, to overcome this defciency,
GP Model-2 is formulated with only two priority goals (minimizing least cost and
DMI), and the results obtained revealed that all the goals are completely achieved
with least cost of Rs 101.6090/- on DM basis, which leads to approximately 30 kg of
TMR, amounting to Rs 4.7/- per kg. Goals 1 and 2 are fully achieved, and remain-
ing goals are either underachieved or overachieved. GP Model-3 is also formulated
with the objective of minimizing the cost of the ration by not considering dry mat-
ter as a goal in the objective function. Dry matter is treated as a constraint directly,
from which we observe that on assigning the weights P1 (goal 1: cost), P2 (goal
2: CP), P3 (goal 3: TDN), P4 (goal 4: Ca), P5 (goal 5: P), P6 (goal 6: roughage), P7
(goal 7: concentrate) as 0.9, 0.8, 0.7, 0.6, 0.5, 0.4 and 0.3 and solving the GP Model
using RGA with hybrid function, we obtain: dcost− = 0.0021, dCP+ = 0.4508, dTDN+ =
0.0901, dCa+ = 0.9891, dPh+ = 0.0499, and the rest of the deviational variables dcost+,
dCP−, dTDN−, dCa−, dPh−, dRough−, dRough+, dConc−, dConc+ are zero. We observe that goal 1
is overachieved; goals 2, 3, 4 and 5 are underachieved, while goals 6 and 7 are fully
achieved without any deviation, with a minimum cost of Rs 101.605/- on DM basis.
All three GP models give a wide range of solutions, but the obtained result does not
completely satisfy the decision maker/nutritionist, as the reasons for underachieve-
ment and overachievement targets need to be analysed further, where the choice of
a fnal solution depends upon the decision maker. There is a need for further discus-
sion with a nutritionist for better output.
The results of the study from Chapters 4, 5 and 6 reveal that after keeping the
nutritional requirements of dairy cattle and buffalo at different body conditions,
the integrated linear and simple goal programming model with weighted priority
function was suffcient to optimize the ration cost. The ration cost obtained by this
method was reasonably cheaper than when formulated by only the LP approach.
The requirement of nutrients (CP, TDN, Ca, P, roughage and concentrates) was
met by fne-tuning the individual feedstuffs. The roughage–concentrates and other
constraints maximum–minimum range make the model more rigid, and hence the
binary coded genetic algorithm was not suggested to solve the problem. To over-
come the limitation of the binary genetic algorithm, it was also proved that the
use of the RGA is a better technique for a good quality feed mix to dairy cattle
for better health and milk production over other conventional methods, viz. LP
simplex, GRG nonlinear and EA and hence can also be used for ration formulation
https://t.me/livestockscience
Conclusion 93
to fnd least cost feedstuffs in dairy cattle. All the techniques – viz. RGA without
crossover, RGA without mutation, RGA with crossover and mutation and RGA with
hybrid function – for least cost ration perform equally. RGA without crossover and
RGA without mutation provide near-optimal answers, but solutions seem to get
stuck in local minima; hence, it is proved that RGA with a hybrid function provides
the optimal solution, and this method can also be used for a ration formulation to
fnd least cost feedstuffs for dairy buffalo.
https://t.me/livestockscience
https://t.me/livestockscience
References
95
https://t.me/livestockscience
96 References
https://t.me/livestockscience
References 97
https://t.me/livestockscience
98 References
https://t.me/livestockscience
References 99
https://t.me/livestockscience
100 References
https://t.me/livestockscience
References 101
https://t.me/livestockscience
102 References
https://t.me/livestockscience
Index
Note: Numbers in bold indicate tables and those in italics indicate fgures.
103
https://t.me/livestockscience
104 Index
https://t.me/livestockscience
Index 105
https://t.me/livestockscience
106 Index
https://t.me/livestockscience
Index 107
https://t.me/livestockscience
108 Index
Least Cost Balanced Ration for Non-Pregnant Mitani and Nakayama on, 14
Dairy Buffalo, 74 overcoming limitations of, 14, 33
Least Cost, Deviation, and Constraints ration formulation and, 10–14
Value Solved by Hybrid RGA for Goal Rehman and Romero on, 61, 71
Programming Models, 66 Right-hand-side (RHS) presumptions
Least-cost feed formulation for dairy cattle/ of, 71
cows during pregnancy, 45–68 simplex method, 11, 29, 33, 62, 89, 92
Least Cost Feed Formulation of Dairy simplex method in MS-Excel, 22
Cattle, 49 Linear Programming Problems (LPP) 60,
Least Cost of Ration by Various Techniques 69; see also, nonlinear programming
for Cattle-1 on As Fresh Basis (LP problems (NLPP)
Model), 54 Level-1, 2 and 3 by genetic algorithm
Least Cost of Ration by Various Techniques (GA), 36
for Cattle-1 on DM Basis, 54 Linear programming (LP) techniques, 46
Least Cost of Ration by Various Techniques Livestock census, All India report, 2, 60, 70
for Cattle-2 on As Fresh Basis (LP
Model), 56 M
Least Cost of Ration by Various Techniques
for Cattle-2 on DM Basis (LP Model), 55 Maize, 5; see also, Maize fodder, Maize
Least Cost of Ration by Various Techniques stover
for Cattle-3 on As Fresh Basis, 57 Maize fodder, 47
Least Cost of Ration by Various Techniques least cost and deviation value, hybrid RGA,
for Cattle-3 on DM Basis (LP Model), 56 GP Model-1, 82
Least Cost Ration Formulated by Various least cost and deviation value, hybrid RGA,
Techniques for Non-Pregnant Dairy Buffalo GP Model-2, 83
on “As Fresh Basis” (LP Model), 81 least cost and deviation value, hybrid RGA,
Least Cost Ration Formulated by Various GP Model-3, 84
Techniques for Non-Pregnant Dairy least cost, deviation, and constraints values,
Buffalo on DM Basis (LP Model), 79 hybrid RGA, 66
Legumes, 5, 34, 35 least cost ration, dairy buffalo, ‘as fresh’
Linear interactive and discrete optimizer basis, 81
(LINDO), 13 least cost ration, dairy buffalo, dry matter
Linearity, 11 (DM) basis, 79
Linear programming based software, 13 nutrient content, 6
Linear programming (LP) model, 87 nutrient content on dry matter (DM) basis,
animal feed formulation, history of use in, 12 48, 73
assumptions informing, 11–12 Maize stover, see Maize fodder
cattle-1, 50, 90, 91 Markov process, 2
cattle-2, 50, 90, 91 MATLAB (Matrix-Laboratory), 13, 21,
cattle-3, 51, 90, 91 23–24, 25, 52
computation of, using MS-Excel Solver, ‘fmincon,’ 79
22–25 ‘gaoptimset,’ 65, 77
dairy buffalo, 74, 79, 80, 81 Maximum Milk Yield Using GA (Original
Danzig, George, 10–11 Bounds), 32
goal programming (GP) combined with, Maximum Milk Yield Using GA (Reduced
20, 46 Bounds), 32
goal programming model and, 68, 72 Maximum Milk Yield Using RST and GA
history of, 10–11 (Original Bounds), 32
hybrid genetic analysis (GA) used to Maximum Milk Yield Using RST and GA
solve, 92 (Reduced Bounds), 33
Kuntal’s improvements on model, 68 MCDA, see Multi-criteria decision analysis
limitations in ration formulation, 13–14, MCDM process, see Multi-criteria decision
19, 61 making (MCDM) process
https://t.me/livestockscience
Index 109
https://t.me/livestockscience
110 Index
https://t.me/livestockscience
Index 111
least cost, deviation, and constraints values, Indian buffaloes in Karnataka, RHGA to
hybrid RGA, 66 solve, 86
least cost ration, dairy buffalo, ‘as fresh’ linear programming in, 10–14
basis, 81 RHGA developed to solve, 86, 91
least cost ration, dairy buffalo, dry matter Real coded genetic algorithm (RGA), 87
(DM) basis, 79 as heuristic technique, 58, 59, 87, 90
nutrient content, 6 hybrid, 70, 82, 83, 84
nutrient content on dry matter (DM) basis, with hybrid function, 65–67, 68, 79, 80, 81,
48, 73 86, 92
Priority Values for GP Model-1, 76 least cost feed formulation for dairy cattle
Priority Values for GP Model-2, 76 using, 47–68
Plots, see Representation of plots least cost ration for non-pregnant dairy
Pregnancy of cows and buffalo, see also, fetus buffaloes, 69–87
growth linear programming (LP) model solved by,
energy requirements, 5, 10 46, 89
least cost feed formulation using RGA numerical optimization using, 52
45–68, 87 offspring provided by, three types of, 53
required nutrients as fxed constraint, 5 outline of steps, 52–53
Proportionality, 11 with random number generation (RNG), 54,
Protein, 3; see also, Crude protein (CP) 55, 56, 57, 58
with seeding, 46, 59, 87, 89, 90
Q without seeding, 54, 55, 56, 57, 58, 59, 87
Real coded hybrid genetic algorithm
Queuing theory, 2 (RHGA), 60
development of, 77–78
R effciency and effectiveness of, 86, 91
least cost ration for buffaloes, used to
Ragi straw, 6, 47 formulate, 69–86
least cost and deviation value, hybrid RGA, limits of binary coded GA overcome by, 87
GP Model-1, 82 ration formulation problem solved by, 86, 91
least cost and deviation value, hybrid RGA, total mixed ration, used to cut down, 70
GP Model-2, 83 Representation of plots, 77–78
least cost and deviation value, hybrid RGA, RGA, see Real coded genetic algorithm
GP Model-3, 84 RHGA, see Real coded hybrid genetic
least cost, deviation, and constraints values, algorithm
hybrid RGA, 66 RNG, see Random number generation
least cost ration, dairy buffalo, ‘as fresh’ RST, see Controlled random search technique
basis, 81 Roughage, defnition of 5; see also, dry
least cost ration, dairy buffalo, dry matter roughage; green roughage
(DM) basis, 79
nutrient content, 6 S
nutrient content on dry matter (DM) basis,
48, 73 Salt, 47
Random number generation concentrate mixtures, 28
real coded genetic algorithm (RGA) with, least cost and deviation value, hybrid RGA,
54, 55, 56, 57, 58 GP Model-1, 82
Ration formulation problem least cost and deviation value, hybrid RGA,
advantages of using spreadsheets to GP Model-2, 83
solve, 21 least cost and deviation value, hybrid RGA,
binary coded genetic algorithm to solve, GP Model-3, 85
27–43 least cost, deviation, and constraints values,
goal-programming model to solve, 33–43 hybrid RGA, 66
https://t.me/livestockscience
112 Index
least cost ration, dairy buffalo, ‘as fresh’ basis, 81 least cost and deviation value, hybrid RGA,
least cost ration, dairy buffalo, dry matter GP Model-3, 84
(DM) basis, 80 least cost, deviation, and constraints values,
nutrient content, 6 hybrid RGA, 66
nutrient content on DM basis, 48, 73 least cost ration, dairy buffalo, ‘as fresh’ basis, 81
Schematic Diagram Representing least cost ration, dairy buffalo, dry matter
Methodology Followed for Least Cost (DM) basis, 79
Feed Formulation of Dairy Cattle, 49 nutrient content, 6
Selected feedstuffs 47; see also, feedstuffs nutrient content on dry matter (DM) basis,
sequential quadratic programming (SQP), 79 48, 73
Simple objective, 11 Stochastic programming
Simplex linear programming (LP), 53, 62 defnition of, 17
MS-Excel Solver, 22 feed formulation method using, 17
Simplex method, 11 linear, 12
controlled random search technique using, 16 WINFEED, 13
FEEDMU, 13 Stochastic techniques, 16
least cost of rations, results per 54–57 Survival of the fttest, principle of, 30
linear programming (LP) problems solved
by, 15, 29, 33, 46, 60 T
Soda bicarbonate, 6
Solution for NLP Model by GA for Original TDN see Total digestible nutrients
Bounds (Test Problem-1), 30 Total digestible nutrients (TDN), 4, 87, 89, 90
Solution for NLP Model by GA for Reduced in common feeds and fodder, 6
Bounds (Test Problem-2), 31 concentrates high in, 5
Solution Set of LPP for Level-1, 2 and 3 by evolutionary algorithm (EA) method, 58
GA, 36 least cost and deviation value, hybrid RGA,
Solution Set of the Nonlinear Programming GP Model-1, 83
Problem (NLPP) for Level-1, 2 and 3 by least cost and deviation value, hybrid RGA,
GA, 37 GP Model-2, 84
Sorghum least cost and deviation value, hybrid RGA,
Co-FS 29, 47, 48, 66, 73, 79, 81, 82, 83 GP Model-3, 85
as dry roughage, 5 least cost for dairy buffalo, 80
least cost and deviation value, hybrid RGA, least cost for cattle-1 on a DM basis, 54, 59, 62
GP Model-1, 82 least cost for cattle-2 on a DM basis, 55, 62
least cost and deviation value, hybrid RGA, least cost for cattle-3 on a DM basis, 57, 62
GP Model-2, 83 linear relationship between milk yield of
least cost and deviation value, hybrid RGA, cows and, 28
GP Model-3, 84 nutrient content of feedstuffs on a dry matter
least cost, deviation, and constraints values, (DM) basis, 48, 73
hybrid RGA, 66 nutrient requirement for buffalo, 72
least cost ration, dairy buffalo, ‘as fresh’ priority values for GP model-1, 76
basis, 81 software program developed to address, 45
least cost ration, dairy buffalo, dry matter Total mixed ration (TMR) cost
(DM) basis, 79 cutting down (reducing) while retaining
multi-cut, 47 nutrients, 60, 70
nutrient content, 6 on dry matter basis, 59, 90
nutrient content on dry matter (DM) basis, farmer’s point of view on, 85–85
48, 73 on fresh basis, 55, 56, 57, 81, 83, 84, 85, 92
Soya de-oiled cake (DOC), 47
least cost and deviation value, hybrid RGA, U
GP Model-1, 82
least cost and deviation value, hybrid RGA, Urea, 6
GP Model-2, 83 Uttar Pradesh, 2, 61
https://t.me/livestockscience
Index 113
e
nc
ie
k sc
oc
st
li ve
e/
t.m
s ://
tp
ht
https://t.me/livestockscience
https://t.me/livestockscience